@neverinfamous/postgres-mcp 1.1.0 → 1.3.0

package/dist/constants/ServerInstructions.d.ts

@@ -9,5 +9,5 @@
  * - Removed redundant alias documentation (already in tool schemas)
  * - Semantic tagging for high-signal guidance
  */
-export declare const SERVER_INSTRUCTIONS = "# postgres-mcp Code Mode\n\n## \u26A0\uFE0F Critical Gotchas\n\n1. **Transactions**: `pg.transactions.execute({statements: [{sql: \"...\"}]})` auto-commits on success, auto-rollbacks on error. To join existing transaction: `{transactionId: txId, statements: [...]}` (no auto-commit, caller controls)\n2. **pg_write_query**: \u26D4 Throws for SELECT\u2014use `pg_read_query` for SELECT statements\n3. **pg_upsert/pg_create_table**: `schema.table` format auto-parses (e.g., `'myschema.users'` \u2192 schema: 'myschema', table: 'users')\n4. **pg_create_table columns**: `notNull`, `defaultValue` (string literals auto-quoted; numbers/booleans auto-coerced; `now()` \u2192 `CURRENT_TIMESTAMP`), `check`, `references` (object or string `\"table(column)\"` syntax)\n5. **pg_create_table constraints**: `constraints` array only accepts `{type: 'unique'|'check'}`. Primary keys: use `column.primaryKey` or top-level `primaryKey: ['col1', 'col2']`\n6. **pg_create_index expression**: Columns can be expressions like `LOWER(name)` or `UPPER(email)`\u2014auto-detected. \u26A0\uFE0F Cast syntax (`::`) requires raw SQL via `pg_write_query`\n7. **pg_list_objects type**: Use `type` (singular string) or `types` (array). Auto-converts: `{type: 'table'}` \u2261 `{types: ['table']}`\n8. **pg_object_details**: Accepts: `name`, `objectName`, `object`, or `table`. Use `type`/`objectType` for type hint (supports: table, view, materialized_view, partitioned_table, function, sequence, index)\n9. **pg_exists optional WHERE**: `where`/`condition`/`filter` is optional. Without it, checks if table has any rows\n10. **pg_describe_table**: Returns columns, foreignKeys, primaryKey\u2014use `pg_get_indexes` separately for index details\n11. **pg_vector_insert updateExisting**: Uses direct UPDATE (avoids NOT NULL constraint issues vs INSERT mode)\n12. **pg_get_indexes without table**: Returns ALL database indexes (potentially large). Use `table` param for specific table\n13. **pg_upsert/pg_batch_insert RETURNING**: `returning` param must be array of column names: `[\"id\", \"name\"]`. \u26D4 `\"*\"` wildcard not supported\n14. **Small tables**: Optimizer correctly uses Seq Scan for <1000 rows\u2014this is expected behavior\n\n## \uD83D\uDD04 Response Structures\n\n| Tool | Returns | Notes |\n|------|---------|-------|\n| `pg_read_query` | `{rows, rowCount, fields?}` | `fields` contains column metadata (name, dataTypeID) |\n| `pg_write_query` | `{rowsAffected, affectedRows, rows?}` | `rows` only with RETURNING clause. \u26D4 Throws for SELECT |\n| `pg_upsert` | `{operation, rowsAffected, rowCount, rows?}` | `operation: 'insert'|'update'`. `rows` only with RETURNING clause |\n| `pg_batch_insert` | `{rowsAffected, affectedRows, insertedCount, rows?}` | Empty objects use DEFAULT VALUES. \u26A0\uFE0F BIGINT > 2^53 loses precision |\n| `pg_count` | `{count: N}` | Use `params` for placeholders: `where: 'id=$1', params: [5]`. DISTINCT: use `pg_read_query` |\n| `pg_exists` | `{exists: bool, mode, hint?}` | `params` for placeholders. `mode: 'filtered'|'any_rows'` |\n| `pg_get_indexes` | `{indexes, count, totalCount?}` | Default `limit: 100` without `table`. Use `schema`/`limit` to filter. Index objects have `name`, `type`, `columns` |\n| `pg_list_objects` | `{objects, count, totalCount, byType}` | Use `limit` to cap results, `type`/`types` to filter |\n| `pg_object_details` | `{name, schema, type, returnType?, ...}` | Functions: `returnType` alias. Views/Mat. views: `definition` |\n| `pg_analyze_db_health` | `{cacheHitRatio: {ratio, heap, index, status}}` | `ratio` = primary numeric %. `bloat` available |\n| `pg_describe_table` | `{columns, indexes, constraints, foreignKeys}` | Columns include `notNull` (alias for `!nullable`), `foreignKey`. `constraints` includes PK, UNIQUE, CHECK, NOT NULL. \u26A0\uFE0F `rowCount: -1` = no statistics (run ANALYZE) |\n| `pg_analyze_query_indexes` | `{plan, issues, recommendations}` | `verbosity`: 'summary' (default) or 'full'. Summary mode returns condensed plan |\n| `pg_list_tables` | `{tables, count}` | Use `schema` to filter, `limit` to cap results |\n| List operations | `{items, count}` | Access via `result.tables`, `result.views`, etc. |\n| `pg_jsonb_agg groupBy` | `{result: [{group_key, items}], count, grouped: true}` | Without groupBy: `{result: [...], count, grouped: false}` |\n| `pg_vector_aggregate` | `{average_vector, count}` or `{groups: [{group_key, average_vector, count}]}` | Without/with `groupBy` |\n\n## API Mapping\n\n`pg_group_action` \u2192 `pg.group.action()` (group prefixes dropped: `pg_jsonb_extract` \u2192 `pg.jsonb.extract()`)\n\n**Top-Level Core Aliases**: All starter tools available directly: `pg.readQuery()`, `pg.writeQuery()`, `pg.listTables()`, `pg.describeTable()`, `pg.createTable()`, `pg.dropTable()`, `pg.count()`, `pg.exists()`, `pg.upsert()`, `pg.batchInsert()`, `pg.truncate()`, `pg.createIndex()`, `pg.dropIndex()`, `pg.getIndexes()`, `pg.listObjects()`, `pg.objectDetails()`, `pg.listExtensions()`, `pg.analyzeDbHealth()`, `pg.analyzeQueryIndexes()`, `pg.analyzeWorkloadIndexes()`\n\n**Positional args work**: `readQuery(\"SELECT...\")`, `exists(\"users\", \"id=1\")`, `createIndex(\"users\", [\"email\"])`\n\n**Discovery**: `pg.help()` returns `{group: methods[]}` mapping (e.g., `{core: ['readQuery', ...], jsonb: [...]}`). `pg.core.help()`, `pg.jsonb.help()` for group-specific methods.\n\n## Format Auto-Resolution\n\n- **Schema.Table**: `'public.users'` auto-parses to `{schema: 'public', table: 'users'}`\n- **JSONB Paths**: Both `'a.b.c'` (string) and `['a','b','c']` (array) work. Use array for literal dots: `[\"key.with.dots\"]`\n- **Aliases**: Common parameter variations resolve automatically (e.g., `query`/`sql`, `table`/`tableName`)\n\n---\n\n## Vector Tools\n\n\u26A0\uFE0F **Large Vectors**: Direct MCP tool calls may truncate vectors >256 dimensions due to JSON-RPC message size limits. For vectors \u2265256 dimensions (e.g., OpenAI 1536-dim, local 384-dim), use Code Mode: `await pg.vector.search({table, column, vector, limit})`\n\n- `pg_vector_search`: Supports `schema.table` format (auto-parsed). Returns `{results: [...], count, metric}`. Use `select: [\"id\", \"name\"]` to include identifying columns. Without select, only returns distance. `filter` = `where`. \u26A0\uFE0F Vectors read from DB are strings\u2014parse before passing: `vec.replace(/^\\[|\\]$/g, '').split(',').map(Number)`\n- `pg_vector_insert`: Supports `schema.table` format (auto-parsed). Use `updateExisting` + `conflictColumn` + `conflictValue` for UPDATE mode. `additionalColumns` is applied in both INSERT and UPDATE modes\n- `pg_vector_batch_insert`: `vectors` expects `[{vector: [...], data?: {...}}]` objects, not raw arrays\n- `pg_vector_normalize`: Returns `{normalized: [...], magnitude: N}`. Note: `magnitude` is the **original** vector length (not 1)\n- `pg_vector_aggregate`: Supports `schema.table` format (auto-parsed). \u26D4 Validates column is vector type. Returns `{average_vector: {preview, dimensions, truncated}, count}` or `{groups: [{group_key, average_vector, count}]}` with groupBy. \u26A0\uFE0F `groupBy` only supports simple column names (not expressions)\n- `pg_vector_dimension_reduce`: Direct mode returns `{reduced: [...], originalDimensions, targetDimensions}`. Table mode returns `{rows: [{id, original_dimensions, reduced}], processedCount, summarized}`. Default `summarize: true` in table mode returns compact `{preview, dimensions, truncated}` format. Use `summarize: false` for full vectors\n- `pg_vector_distance`: Calculate distance between two vectors. `metric`: 'l2' (default), 'cosine', 'inner_product'. Returns `{distance, metric}`\n- `pg_vector_cluster`: `clusters` = `k`. Returns centroids with `{preview, dimensions, truncated}` format for large vectors (>10 dims)\u2014use `pg_vector_distance` to assign rows\n- `pg_vector_create_index`: Use `type` (or alias `method`) with values 'ivfflat' or 'hnsw'. IVFFlat: `lists` param. HNSW: `m`, `efConstruction` params\n- `pg_vector_performance`: Auto-generates testVector from first row if omitted. Returns `testVectorSource: 'auto-generated'|'user-provided'`\n- `pg_vector_validate`: Returns `{valid: bool, vectorDimensions}`. Empty vector `[]` returns `{valid: true, vectorDimensions: 0}`\n- \u26D4 `pg_vector_embed`: Demo only (hash-based). Use OpenAI/Cohere for production.\n- `pg_hybrid_search`: Supports `schema.table` format (auto-parsed). Combines vector similarity and full-text search with weighted scoring. Code mode alias: `pg.hybridSearch()` \u2192 `pg.vector.hybridSearch()`\n- \uD83D\uDCDD **Error Handling**: Vector tools return `{success: false, error: \"...\", suggestion: \"...\"}` for validation/semantic errors (dimension mismatch, non-vector column, table not found). Check `success` field before processing results.\n\n## JSONB Tools\n\n- `pg_jsonb_extract`: Returns null if path doesn't exist\n- `pg_jsonb_insert`: Index -1 inserts BEFORE last element; use `insertAfter: true` to append. \u26A0\uFE0F Use array format `[-1]` not string `\"[-1]\"` for negative indices\n- `pg_jsonb_set`: `createMissing=true` creates full nested paths; initializes NULL columns to `{}`. Empty path (`''` or `[]`) replaces entire column value\n- `pg_jsonb_strip_nulls`: \u26A0\uFE0F Requires `where`/`filter` clause\u2014write operations must be targeted. Use `preview: true` to see changes first\n- `pg_jsonb_agg`: Supports AS aliases in select: `[\"id\", \"metadata->>'name' AS name\"]`. \u26A0\uFE0F `->>` returns text\u2014use `->` to preserve JSON types\n- `pg_jsonb_object`: Use `data`, `object`, or `pairs` parameter: `{data: {name: \"John\", age: 30}}`. Returns `{object: {...}}`\n- `pg_jsonb_normalize`: `flatten` doesn't descend into arrays; `keys` returns text (use `pairs` for JSON types)\n- `pg_jsonb_stats`: Returns column-level statistics. `topKeysLimit` controls key count (default: 20). \u26A0\uFE0F `typeDistribution` null type = SQL NULL columns (entire column NULL, not JSON `null` literal). Use `sqlNullCount` for explicit count\n- \u26D4 **Object-only tools**: `diff`, `merge`, `keys`, `indexSuggest`, `securityScan`, `stats`\u2014topKeys require JSONB objects, throw descriptive errors for arrays\n- \u26D4 **Array-only tools**: `insert`\u2014requires JSONB arrays, throws errors for objects\n- \uD83D\uDCDD `normalize` modes: `pairs`/`keys`/`flatten` for objects; `array` for arrays\n\n**Top-Level Aliases**: `pg.jsonbExtract()`, `pg.jsonbSet()`, `pg.jsonbInsert()`, `pg.jsonbDelete()`, `pg.jsonbContains()`, `pg.jsonbPathQuery()`, `pg.jsonbAgg()`, `pg.jsonbObject()`, `pg.jsonbArray()`, `pg.jsonbKeys()`, `pg.jsonbStripNulls()`, `pg.jsonbTypeof()`, `pg.jsonbValidatePath()`, `pg.jsonbMerge()`, `pg.jsonbNormalize()`, `pg.jsonbDiff()`, `pg.jsonbIndexSuggest()`, `pg.jsonbSecurityScan()`, `pg.jsonbStats()`\n\n\n## Stats Tools\n\n- All stats tools support `schema.table` format (auto-parsed, embedded schema takes priority over explicit `schema` param)\n- `timeSeries`: Both `timeColumn` (must be timestamp/date) and `valueColumn` (must be numeric) are validated upfront with clear error messages. Aliases: `time`\u2192`timeColumn`, `value`\u2192`valueColumn`. `interval` accepts: `second`, `minute`, `hour`, `day`, `week`, `month`, `year` (keywords, PostgreSQL format, or plurals). Default `limit: 100` time buckets. Use `limit: 0` for no limit. Returns `truncated` and `totalCount` indicators when default limit is applied. **groupBy payloads**: Default `groupLimit: 20` groups. Returns `truncated` + `totalGroupCount` when groups are limited. Use `groupLimit: 0` for all groups\n- `correlation`: Use `column1`/`column2` or aliases `x`/`y` for column names\n- `distribution`: Returns `skewness`, `kurtosis` (excess). `buckets` must be > 0. **groupBy payloads**: Default `groupLimit: 20` groups (prevents large payloads with many histogram buckets per group). Returns `truncated` + `totalGroupCount` when groups are limited. Use `groupLimit: 0` for all groups\n- `sampling`: Defaults to `random` method with 20 rows (optimized for LLM context). `sampleSize` always takes precedence over `percentage`. \u26A0\uFE0F `percentage` param only works with `bernoulli`/`system` methods\u2014ignored for default `random` method. Default limit of 100 rows applied to `bernoulli`/`system` with `percentage` to prevent large payloads. Returns `truncated` and `totalSampled` when TABLESAMPLE returns more rows than limit\n- `percentiles`: Accepts 0-1 or 0-100 (auto-normalized). \u26A0\uFE0F Use consistent scale\u2014mixing (e.g., `[0.1, 50]`) produces unexpected keys and returns a `warning` field explaining the issue. Empty array \u2192 defaults [0.25, 0.5, 0.75]\n- `hypothesis`: Returns nested `results` object containing `pValue` (two-tailed), `testStatistic`, `interpretation`, `sampleMean`, `sampleStdDev`. Access via `hyp.results.pValue`. Use `populationStdDev` for z-test, otherwise defaults to t-test\n- `regression`: Use `xColumn`/`yColumn`, aliases `x`/`y`, or `column1`/`column2` (for consistency with correlation). Returns nested `regression` object containing `slope`, `intercept`, `rSquared`, `equation`, `avgX`, `avgY`, `sampleSize`. Access via `reg.regression.slope`\n- \u26A0\uFE0F WARNING: `sampling` with `system` method unreliable for small tables\u2014use `bernoulli` or `random`\n\n**Top-Level Aliases**: `pg.descriptive()`, `pg.percentiles()`, `pg.correlation()`, `pg.regression()`, `pg.timeSeries()`, `pg.distribution()`, `pg.hypothesis()`, `pg.sampling()`\n\n## Performance Tools\n\nCore (20 methods): `explain()`, `explainAnalyze()`, `explainBuffers()`, `indexStats()`, `tableStats()`, `statStatements()`, `statActivity()`, `locks()`, `bloatCheck()`, `cacheHitRatio()`, `seqScanTables()`, `indexRecommendations()`, `queryPlanCompare()`, `baseline()`, `connectionPoolOptimize()`, `partitionStrategySuggest()`, `unusedIndexes()`, `duplicateIndexes()`, `vacuumStats()`, `queryPlanStats()`\n\nWrappers (3): `blockingQueries()`\u2192`locks({showBlocked:true})`, `longRunningQueries({ seconds | minDuration }?)` filters by duration (returns `statActivity` format), `analyzeTable({ table })` runs ANALYZE (accepts `schema.table` format)\n\n- `explain({ sql, format?, params? })`: Supports `format: 'text'|'json'|'yaml'|'xml'`. Default: text. Use `params: [value]` for `$1, $2` placeholders\n- `explainAnalyze({ sql, format?, params? })`: Same format/params options as explain\n- `explainBuffers({ sql, params? })`: Always returns JSON format (includes buffer statistics)\n- `indexRecommendations({ sql?, params? })`: Pass `params: [value]` for parameterized queries (e.g., `sql: 'SELECT * FROM orders WHERE id = $1', params: [5]`)\n- `queryPlanCompare({ query1, query2, params1?, params2? })`: Compare two query plans. Use `params1`/`params2` for parameterized queries\n- `partitionStrategySuggest({ table })`: Accepts `schema.table` format (auto-parsed) or separate `table` + `schema` params\n- \u26A0\uFE0F **Data Type Awareness**: Query literals must match column types exactly\u2014`WHERE sensor_id = 1` (integer), not `'sensor_1'` (string)\n\nAliases: `cacheStats`\u2192`cacheHitRatio`, `queryStats`\u2192`statStatements`, `activity`\u2192`statActivity`, `vacuum`\u2192`vacuumStats`, `indexUsage`\u2192`indexStats`, `bloatEstimate`/`bloat`\u2192`bloatCheck`, `runningQueries`\u2192`longRunningQueries`\n\n\uD83D\uDCE6 **AI-Optimized Payloads**: Tools return limited results by default to reduce context size:\n- `indexStats({ limit? })`: Default 50 rows. Returns `truncated: true` + `totalCount` when limited. Use `limit: 0` for all\n- `tableStats({ limit? })`: Default 50 rows. Returns `truncated: true` + `totalCount` when limited. Use `limit: 0` for all\n- `vacuumStats({ limit? })`: Default 50 rows. Same truncation indicators. Use `limit: 0` for all\n- `statStatements({ limit?, orderBy? })`: Default 20 rows. Returns `truncated: true` + `totalCount` when limited. Use `limit: 0` for all\n- `unusedIndexes({ limit?, summary? })`: Default 20 rows. Use `summary: true` for aggregated stats by schema\n- `queryPlanStats({ limit?, truncateQuery? })`: Default 20 rows, queries truncated to 100 chars. Use `truncateQuery: 0` for full text\n\n\uD83D\uDCCD **Code Mode Note**: `pg_performance_baseline` \u2192 `pg.performance.baseline()` (not `performanceBaseline`). `indexRecommendations` accepts `query` alias for `sql`\n\n**Top-Level Aliases**: `pg.explain()`, `pg.explainAnalyze()`, `pg.cacheHitRatio()`, `pg.indexStats()`, `pg.tableStats()`, `pg.indexRecommendations()`, `pg.bloatCheck()`, `pg.vacuumStats()`, `pg.unusedIndexes()`, `pg.duplicateIndexes()`, `pg.seqScanTables()`\n\n## Monitoring Tools\n\nCore: `databaseSize()`, `tableSizes()`, `connectionStats()`, `showSettings()`, `capacityPlanning()`, `uptime()`, `serverVersion()`, `recoveryStatus()`, `replicationStatus()`, `resourceUsageAnalyze()`, `alertThresholdSet()`\n\n- `databaseSize()`: Returns `{bytes: number, size: string}`. Optional `database` param for specific db\n- `tableSizes({ limit?, schema? })`: Default limit 50. Returns `{tables: [...], count, truncated?, totalCount?}`. `truncated: true` + `totalCount` when limited. Use `limit: 0` for all\n- `connectionStats()`: Returns `{byDatabaseAndState, totalConnections: number, maxConnections: number}`\n- `showSettings({ setting?, limit? })`: Default limit 50 when no pattern. Returns `{settings: [...], count, truncated?, totalCount?}`. Accepts `pattern`, `setting`, or `name`. Exact names auto-match; `%` for LIKE patterns\n- `capacityPlanning({days: 90})`: `days` = `projectionDays`. Returns `{current, growth, projection, recommendations}` with numeric fields. \u26D4 Negative days rejected\n- `uptime()`: Returns `{start_time: string, uptime: {days, hours, minutes, seconds, milliseconds}}`\n- `serverVersion()`: Returns `{full_version: string, version: string, version_num: number}`\n- `recoveryStatus()`: Returns `{in_recovery: boolean, last_replay_timestamp: string|null}`\n- `replicationStatus()`: Returns `{role: 'primary'|'replica', replicas: [...]}` for primary, or `{role: 'replica', replay_lag, ...}` for replica\n- `resourceUsageAnalyze()`: Returns `{backgroundWriter, checkpoints, connectionDistribution, bufferUsage, activity, analysis}` with all counts as numbers\n- `alertThresholdSet({metric?: 'connection_usage'})`: Returns recommended thresholds. \u26D4 Invalid metric throws validation error. Valid metrics: connection_usage, cache_hit_ratio, replication_lag, dead_tuples, long_running_queries, lock_wait_time\n\n\uD83D\uDCE6 **AI-Optimized Payloads**: Tools return limited results by default to reduce context size:\n- `tableSizes({ limit? })`: Default 50 rows. Returns `truncated: true` + `totalCount` when limited. Use `limit: 0` for all\n- `showSettings({ limit? })`: Default 50 rows when no pattern specified. Use `limit: 0` for all or specify a pattern\n\nAliases: `tables`\u2192`tableSizes`, `connections`\u2192`connectionStats`, `settings`/`config`\u2192`showSettings`, `alerts`/`thresholds`\u2192`alertThresholdSet`\n\n**Top-Level Aliases**: `pg.databaseSize()`, `pg.tableSizes()`, `pg.connectionStats()`, `pg.serverVersion()`, `pg.uptime()`, `pg.showSettings()`, `pg.recoveryStatus()`, `pg.replicationStatus()`, `pg.capacityPlanning()`, `pg.resourceUsageAnalyze()`, `pg.alertThresholdSet()`\n\n## Admin Tools\n\nCore: `vacuum()`, `vacuumAnalyze()`, `analyze()`, `reindex()`, `cluster()`, `setConfig()`, `reloadConf()`, `resetStats()`, `cancelBackend()`, `terminateBackend()`\n\n- All admin tools support `schema.table` format (auto-parsed, embedded schema takes priority over explicit `schema` param)\n- `vacuum({ table?, full?, analyze?, verbose? })`: Without `table`, vacuums ALL tables. `verbose` output goes to PostgreSQL server logs\n- `reindex({ target, name?, concurrently? })`: Targets: 'table', 'index', 'schema', 'database'. `database` target defaults to current db when `name` omitted\n- `cluster()`: Without args, re-clusters all previously-clustered tables. With args, requires BOTH `table` AND `index`\n- `setConfig({ name, value, isLocal? })`: `isLocal: true` applies only to current transaction\n- `cancelBackend({ pid })`: Graceful query cancellation\u2014returns `{success: false}` for invalid PID (no error thrown)\n- `terminateBackend({ pid })`: Forceful connection termination\u2014use with caution\n\nAliases: `tableName`\u2192`table`, `indexName`\u2192`index`, `param`/`setting`\u2192`name`, `processId`\u2192`pid`\n\n**Top-Level Aliases**: `pg.vacuum()`, `pg.vacuumAnalyze()`, `pg.analyze()`, `pg.reindex()`, `pg.cluster()`, `pg.setConfig()`, `pg.reloadConf()`, `pg.resetStats()`, `pg.cancelBackend()`, `pg.terminateBackend()`\n\n## Backup Tools\n\nCore: `dumpTable()`, `dumpSchema()`, `copyExport()`, `copyImport()`, `createBackupPlan()`, `restoreCommand()`, `physical()`, `restoreValidate()`, `scheduleOptimize()`\n\nResponse Structures:\n- `dumpTable`: `{ddl, type, note, insertStatements?}` \u2014 `insertStatements` only with `includeData: true` (separate field from `ddl`)\n- `copyExport`: `{data, rowCount, truncated?, limit?}` \u2014 `data` contains CSV/text content. `truncated: true` + `limit` when rows returned equals applied limit (indicating more rows likely exist)\n- `copyImport`: `{command, stdinCommand, notes}` \u2014 Both file and stdin COPY commands\n- `createBackupPlan`: `{strategy: {fullBackup, walArchiving}, estimates}`\n- `restoreCommand`: `{command, warnings?, notes}` \u2014 Warnings when `database` omitted\n- `restoreValidate`: `{validationSteps: [{step, name, command?, commands?, note?}], recommendations}` \u2014 Note: `note` field only for pg_dump default type\n- `physical`: `{command, notes, requirements}`\n- `scheduleOptimize`: `{analysis, recommendation, commands}`\n\n\uD83D\uDCE6 **AI-Optimized Payloads**: `copyExport` limits results to 500 rows by default to prevent large payloads. Use `limit: 0` for all rows, or specify a custom limit.\n\n- `pg_copy_export`: Use `query`/`sql` OR `table`. \u26A0\uFE0F If both provided, `query` takes precedence with warning. Supports `schema.table` format (auto-parsed, takes priority over `schema` param). Format: `csv` (default, comma-delimited), `text` (tab-delimited). Both formats support `header: true` (default). \u26D4 `binary` not supported via MCP\u2014use `pg_dump_schema` for binary exports. Default `limit: 500` (use `0` for all rows). Optional `delimiter` to customize\n- `pg_dump_table`: Returns `ddl` + `insertStatements` when `includeData: true`. Supports sequences (`type: 'sequence'`), views (`type: 'view'`), and partitioned tables (`type: 'partitioned_table'` with `PARTITION BY` clause). **PRIMARY KEYS, INDEXES, CONSTRAINTS NOT included**\u2014use `pg_get_indexes`/`pg_get_constraints`. Supports `schema.table` format\n- `pg_dump_schema`: Generates pg_dump command. Optional `schema`, `table`, `filename`\n- `pg_copy_import`: Generates COPY FROM command. Supports `schema.table` format (auto-parsed, takes priority over `schema` param). `columns` array, `filePath`, `format`, `header`, `delimiter`\n- `pg_restore_command`: Include `database` parameter for complete command. Optional `schemaOnly`, `dataOnly`\n- `pg_create_backup_plan`: Generates backup strategy with cron schedule. `frequency`: 'hourly'|'daily'|'weekly', `retention` count\n- `pg_backup_physical`: Generates pg_basebackup command. `format`: 'plain'|'tar', `checkpoint`: 'fast'|'spread', `compress`: 0-9\n- `pg_restore_validate`: Generates validation commands. `backupType`: 'pg_dump' (default)|'pg_basebackup'\n- `pg_backup_schedule_optimize`: Analyzes database activity patterns and recommends optimal backup schedule\n\n**Top-Level Aliases**: `pg.dumpTable()`, `pg.dumpSchema()`, `pg.copyExport()`, `pg.copyImport()`, `pg.createBackupPlan()`, `pg.restoreCommand()`, `pg.restoreValidate()`, `pg.physical()`, `pg.backupPhysical()`, `pg.scheduleOptimize()`, `pg.backupScheduleOptimize()`\n\n## Text Tools\n\nDefaults: `threshold`=0.3 (use 0.1-0.2 for partial), `maxDistance`=3 (use 5+ for longer strings)\n\n- All text tools support `schema.table` format (auto-parsed, embedded schema takes priority over explicit `schema` param)\n- `pg_text_search`: Supports both `column` (singular string) and `columns` (array). Either is valid\u2014`column` auto-converts to array\n- `pg_trigram_similarity` vs `pg_similarity_search`: Both use pg_trgm. First filters by threshold; second uses set_limit() with %\n- `pg_fuzzy_match`: Levenshtein returns distance (lower=better). Soundex/metaphone return phonetic codes (exact match only). \u26D4 Invalid `method` values throw error with valid options\n- `pg_text_normalize`: Removes accents only (unaccent). Does NOT lowercase/trim\n- \uD83D\uDCCD **Table vs Standalone**: `normalize`, `sentiment`, `toVector`, `toQuery`, `searchConfig` are standalone (text input only). For phonetic matching: use `pg_fuzzy_match` with `method: 'soundex'|'metaphone'` (direct MCP), or `pg.text.soundex()`/`pg.text.metaphone()` (Code Mode convenience wrappers that call fuzzyMatch internally)\n\n**Top-Level Aliases**: `pg.textSearch()`, `pg.textRank()`, `pg.textHeadline()`, `pg.textNormalize()`, `pg.textSentiment()`, `pg.textToVector()`, `pg.textToQuery()`, `pg.textSearchConfig()`, `pg.textTrigramSimilarity()`, `pg.textFuzzyMatch()`, `pg.textLikeSearch()`, `pg.textRegexpMatch()`, `pg.textCreateFtsIndex()`\n\n\n## Schema Tools\n\nCore: `listSchemas()`, `createSchema()`, `dropSchema()`, `listViews()`, `createView()`, `dropView()`, `listSequences()`, `createSequence()`, `dropSequence()`, `listFunctions()`, `listTriggers()`, `listConstraints()`\n\nResponse Structures:\n- `listSchemas()`: `{schemas: string[], count}`\n- `listViews({ includeMaterialized?, truncateDefinition?, limit? })`: `{views: [{schema, name, type, definition, definitionTruncated?}], count, hasMatViews, truncatedDefinitions?, truncated, note?}`. Default `limit: 50` (use `0` for all). Default `truncateDefinition: 500` chars (use `0` for full definitions). `truncated` always included (`true`/`false`)\n- `listSequences({ schema? })`: `{sequences: [{schema, name, owned_by}], count}`. Note: `owned_by` omits `public.` prefix for sequences in public schema (e.g., `users.id` not `public.users.id`)\n- `listFunctions({ schema?, limit?, exclude? })`: `{functions: [{schema, name, arguments, returns, language, volatility}], count, limit, note?}`\n- `listTriggers({ schema?, table? })`: `{triggers: [{schema, table_name, name, timing, events, function_name, enabled}], count}`\n- `listConstraints({ schema?, table?, type? })`: `{constraints: [{schema, table_name, name, type, definition}], count}`. Type codes: `p`=primary_key, `f`=foreign_key, `u`=unique, `c`=check\n- `dropSchema/dropView/dropSequence`: All return `{existed: true/false}` to indicate if object existed before drop\n- `createSchema/createSequence` (with `ifNotExists`) and `createView` (with `orReplace`): Return `{alreadyExisted: true/false}` to indicate if object existed before creation\n\n- `pg_create_view`: Supports `schema.name` format (auto-parsed). Use `orReplace: true` for CREATE OR REPLACE. `checkOption`: 'cascaded', 'local', 'none'. \u26D4 OR REPLACE can add new columns but cannot rename/remove existing ones\u2014PostgreSQL limitation\n- `pg_create_sequence`: Supports `schema.name` format. Parameters: `start`, `increment`, `minValue`, `maxValue`, `cache`, `cycle`, `ownedBy`, `ifNotExists`\n- `pg_list_functions`: Default limit=500. Use `schema: 'public'`, `limit: 2000`, or `exclude: ['postgis']` to filter. \u26A0\uFE0F `exclude` filters by **schema name** AND extension-owned functions. Note: Aggressive `exclude` may return 0 results if all functions belong to excluded extensions\n\n**Discovery**: `pg.schema.help()` returns `{methods: string[], examples: string[]}` object with available methods and usage examples\n\n\n## Partitioning Tools\n\n- `pg_create_partitioned_table`: `partitionBy` case-insensitive. Supports `schema.table` format for `name` (auto-parsed). `primaryKey` accepts array (e.g., `['id', 'event_date']`). \u26D4 `primaryKey`/`unique` must include partition key\u2014throws validation error otherwise\n- `pg_create_partition`: Use `parent`/`table`/`parentTable`. `forValues` is a raw SQL string: `\"FROM ('2024-01-01') TO ('2024-07-01')\"`, `\"IN ('US', 'CA')\"`, `\"WITH (MODULUS 4, REMAINDER 0)\"`. For DEFAULT partition, use `isDefault: true`. Supports `schema.table` format for `parent` (auto-parsed)\n- `pg_attach_partition`/`pg_detach_partition`: Support `schema.table` format for `parent` and `partition` (auto-parsed). For DEFAULT partition, use `isDefault: true` or `forValues: \"DEFAULT\"`\n- `pg_list_partitions`: Default `limit: 50` (use `0` for all). Returns `{partitions, count, truncated, totalCount?}`. Uses `bounds` field (consistent with `pg_partition_info`)\n- `pg_partition_info`: Returns `{tableInfo, partitions, totalSizeBytes}`. Uses `bounds` field\n- Both list/info tools support `schema.table` format (auto-parsed) and accept `table`, `parent`, `parentTable`, or `name` aliases\n- \uD83D\uDCCD Code Mode: `pg.partitioning.create()` = `createPartition`, NOT `createPartitionedTable`\n\n## pg_partman Tools\n\n- `pg_partman_create_parent`: Interval uses PostgreSQL syntax ('1 day', '1 month') NOT keywords ('daily'). `startPartition` accepts 'now' shorthand for current date. Required params: `parentTable`/`table`, `controlColumn`/`control`/`column`, `interval`\n- `pg_partman_run_maintenance`: Without `parentTable`, maintains ALL partition sets. Returns `partial: true` when some tables are skipped. `orphaned` object groups orphaned configs with `count`, `tables`, and cleanup `hint`. `errors` array for other failures\n- `pg_partman_show_config`: Default `limit: 50` (use `0` for all). Returns `truncated` + `totalCount` when limited. `orphaned` flag per config. Supports `schema.table` or plain table name (auto-prefixes `public.`)\n- `pg_partman_show_partitions`: Default `limit: 50` (use `0` for all). Returns `truncated` + `totalCount` when limited. `parentTable` required. Supports `schema.table` format (auto-parsed)\n- `pg_partman_check_default`/`partition_data`: `parentTable` required. Supports `schema.table` format (auto-parsed)\n- `pg_partman_set_retention`: \u26A0\uFE0F **CAUTION: Default is DROP** \u2014 `retentionKeepTable: false` (default) = DROP partitions, `true` = detach only (safer). Pass `retention: null` to disable retention\n- `pg_partman_undo_partition`: `targetTable` MUST exist before calling. Requires both `parentTable` and `targetTable`/`target`\n- `pg_partman_analyze_partition_health`: Default `limit: 50` (use `0` for all). Returns `truncated` + `totalCount` when limited. `summary.overallHealth`: 'healthy'|'warnings'|'issues_found'\n- \uD83D\uDCDD **Schema Resolution**: All partman tools auto-prefix `public.` when no schema specified in `parentTable`\n- \uD83D\uDCDD **Aliases**: `parentTable` accepts `table`, `parent`, `name`. `controlColumn` accepts `control`, `column`. `targetTable` accepts `target`\n\n## pg_stat_kcache Tools\n\nCore: `createExtension()`, `queryStats()`, `topCpu()`, `topIo()`, `databaseStats()`, `resourceAnalysis()`, `reset()`\n\n- `pg_kcache_query_stats`: Default `limit: 50` (use `0` for all). Returns `truncated` + `totalCount` when limited. `orderBy`: 'total_time' (default), 'cpu_time', 'reads', 'writes'. `queryPreviewLength`: chars for query preview (default: 100, max: 500, 0 for full). \u26D4 'calls' NOT valid for orderBy\u2014use `minCalls` param\n- `pg_kcache_resource_analysis`: Default `limit: 50` (use `0` for all). Returns `truncated` + `totalCount` when limited. `minCalls`, `queryPreviewLength` supported. Classifies queries as 'CPU-bound', 'I/O-bound', or 'Balanced'\n- `pg_kcache_top_cpu`: Top CPU-consuming queries. `limit` param (default: 10)\n- `pg_kcache_top_io`: `type`/`ioType` (alias): 'reads', 'writes', 'both' (default). `limit` param (default: 10)\n- `pg_kcache_database_stats`: Aggregated CPU/IO stats per database\n- `pg_kcache_reset`: Resets pg_stat_kcache AND pg_stat_statements statistics\n\n## citext Tools\n\nCore: `createExtension()`, `convertColumn()`, `listColumns()`, `analyzeCandidates()`, `compare()`, `schemaAdvisor()`\n\n- `pg_citext_create_extension`: Enable citext extension (idempotent). Returns `{success, message, usage}`\n- `pg_citext_convert_column`: Supports `schema.table` format (auto-parsed). \u26D4 Only allows text-based columns (text, varchar, character varying)\u2014non-text columns return `{success: false, error, allowedTypes, suggestion}`. When views depend on column, returns `{success: false, dependentViews, hint}`\u2014drop/recreate views manually. `col` alias for `column`. Returns `{previousType}` showing original type\n- `pg_citext_list_columns`: Default `limit: 100` (use `0` for all). Returns `{columns: [{table_schema, table_name, column_name, is_nullable, column_default}], count, totalCount, truncated}`. Optional `schema`, `limit` filters\n- `pg_citext_analyze_candidates`: Default `limit: 50` (use `0` for all). Default `excludeSystemSchemas: true` filters out extension schemas (cron, topology, partman, tiger) when no `schema`/`table` filter specified\u2014use `excludeSystemSchemas: false` to include all. Returns `truncated: true` + `totalCount` when results are limited. Scans tables for TEXT/VARCHAR columns matching common patterns (email, username, name, etc.). Optional `schema`, `table`, `limit`, `excludeSystemSchemas`, `patterns` filters. Returns `{candidates, count, totalCount, truncated, summary: {highConfidence, mediumConfidence}, recommendation, patternsUsed, excludedSchemas?}`\n- `pg_citext_compare`: Test case-insensitive comparison. Returns `{value1, value2, citextEqual, textEqual, lowerEqual, extensionInstalled}`\n- `pg_citext_schema_advisor`: Supports `schema.table` format (auto-parsed). Analyzes specific table. Returns `{table, recommendations: [{column, currentType, previousType?, recommendation, confidence, reason}], summary, nextSteps}`. `tableName` alias for `table`. Already-citext columns include `previousType: \"text or varchar (converted)\"`\n\n**Discovery**: `pg.citext.help()` returns `{methods, methodAliases, examples}` object\n\n## ltree Tools\n\nCore: `createExtension()`, `query()`, `match()`, `subpath()`, `lca()`, `listColumns()`, `convertColumn()`, `createIndex()`\n\n- `pg_ltree_create_extension`: Enable ltree extension (idempotent). Returns `{success, message}`\n- `pg_ltree_query`: Query hierarchical relationships. Supports `schema.table` format (auto-parsed). `mode`/`type`: 'ancestors', 'descendants' (default), 'exact'. Returns `{results, count, path, mode, isPattern}`. \u26A0\uFE0F Validates column is ltree type\u2014returns clear error for non-ltree columns\n- `pg_ltree_match`: Match paths using lquery pattern syntax (`*`, `*{1,2}`, `*.label.*`). Supports `schema.table` format. `pattern`/`lquery`/`query` aliases. Returns `{results, count, pattern}`\n- `pg_ltree_subpath`: Extract portion of ltree path. `offset`/`start`/`from` and `length`/`len` aliases. Negative `offset` counts from end. \u26A0\uFE0F Returns `{success: false, error, pathDepth}` for invalid offset (validated before PostgreSQL call)\n- `pg_ltree_lca`: Find longest common ancestor of multiple paths. Requires `paths` array (min 2). Returns `{longestCommonAncestor, hasCommonAncestor: bool, paths}`\n- `pg_ltree_list_columns`: List all ltree columns in database. Optional `schema` filter. Returns `{columns: [{table_schema, table_name, column_name, is_nullable, column_default}], count}`\n- `pg_ltree_convert_column`: Convert TEXT column to ltree. Supports `schema.table` format. `col` alias for `column`. Returns `{previousType}`. \u26A0\uFE0F When views depend on column, returns `{success: false, dependentViews, hint}`\u2014drop/recreate views manually\n- `pg_ltree_create_index`: Create GiST index on ltree column. Supports `schema.table` format. Auto-generates index name if `indexName` omitted. Returns `{indexName, indexType: 'gist', alreadyExists?}`\n\n**Discovery**: `pg.ltree.help()` returns `{methods, aliases, examples}` object. Top-level aliases available: `pg.ltreeQuery()`, `pg.ltreeMatch()`, etc.\n\n## PostGIS Tools\n\n**Geometry Creation:**\n- `pg_geocode`: Create point geometry from lat/lng. Returns `{geojson, wkt}`. \u26A0\uFE0F Validates bounds: lat \u00B190\u00B0, lng \u00B1180\u00B0\n- `pg_geometry_column`: Add geometry column to table. `ifNotExists` returns `{alreadyExists: true}`\n- `pg_spatial_index`: Create GiST spatial index. Auto-generates name if not provided. `ifNotExists` supported\n\n**Spatial Queries:**\n- `pg_distance`: Find geometries within distance from point. Returns `{results, count}` with `distance_meters`. \u26A0\uFE0F Validates point bounds\n- `pg_bounding_box`: Find geometries within lat/lng bounding box. Use `select` array for specific columns\n- `pg_intersection`: Find geometries intersecting a WKT/GeoJSON geometry. Auto-detects SRID from column\n- `pg_point_in_polygon`: Check if point is within table polygons. Returns `{containingPolygons, count}`. \u26A0\uFE0F Validates point bounds\n\n**Geometry Operations (Table-based):**\n- `pg_buffer`: Create buffer zone around table geometries. Default limit: 50 rows. Default simplify: 10m (set `simplify: 0` to disable). Returns `truncated: true` + `totalCount` when results are truncated. Use `limit: 0` for all rows\n- `pg_geo_transform`: Transform table geometries between SRIDs. Default limit: 50 rows. Returns `truncated: true` + `totalCount` when results are truncated. Use `limit: 0` for all rows. `fromSrid`/`sourceSrid` and `toSrid`/`targetSrid` aliases\n- `pg_geo_cluster`: Spatial clustering (DBSCAN/K-Means). K-Means: If `numClusters` exceeds row count, automatically clamps to available rows with `warning` field. DBSCAN: Returns contextual `hints` array explaining parameter effects (e.g., \"All points formed single cluster\u2014decrease eps\") and `parameterGuide` explaining eps/minPoints trade-offs\n\n**Geometry Operations (Standalone WKT/GeoJSON):**\n- `pg_geometry_buffer`: Create buffer around WKT/GeoJSON. Returns `{buffer_geojson, buffer_wkt, distance_meters}`. Optional `simplify` param (meters) reduces polygon complexity\u2014returns `simplified`, `simplifyTolerance` when applied. \u26A0\uFE0F Returns `warning` if simplify tolerance is too high and geometry collapses to null\n- `pg_geometry_transform`: Transform WKT/GeoJSON between SRIDs. Returns `{transformed_geojson, transformed_wkt, fromSrid, toSrid}`\n- `pg_geometry_intersection`: Compute intersection of two geometries. Returns `{intersects, intersection_geojson, intersection_area_sqm}`. Normalizes SRID (4326) automatically\u2014safe to mix GeoJSON and WKT\n\n**Administration:**\n- `pg_postgis_create_extension`: Enable PostGIS extension (idempotent)\n- `pg_geo_index_optimize`: Analyze spatial indexes. Without `table` param, analyzes all spatial indexes\n\n**Code Mode Aliases:** `pg.postgis.addColumn()` \u2192 `geometryColumn`, `pg.postgis.indexOptimize()` \u2192 `geoIndexOptimize`. Note: `pg.{group}.help()` returns `{methods, aliases, examples}`\n\n## Cron Tools (pg_cron)\n\nCore: `createExtension()`, `schedule()`, `scheduleInDatabase()`, `unschedule()`, `alterJob()`, `listJobs()`, `jobRunDetails()`, `cleanupHistory()`\n\n- `pg_cron_schedule`: Schedule a cron job. `schedule` supports standard cron (`0 5 * * *`) or interval (`1 second` to `59 seconds`). \u26A0\uFE0F Interval syntax only works for 1-59 seconds\u2014for 60+ seconds, use cron syntax (e.g., `* * * * *` for every minute). Use `name`/`jobName` for identification. `command`/`sql`/`query` aliases supported. Note: pg_cron allows duplicate job names; use unique names to avoid confusion when unscheduling\n- `pg_cron_schedule_in_database`: Schedule job in specific database. `database`/`db` aliases. Optional `username`, `active` params\n- `pg_cron_unschedule`: Remove job by `jobId` or `jobName`. If both provided, `jobName` takes precedence (with warning)\n- `pg_cron_alter_job`: Modify existing job. Can change `schedule`, `command`, `database`, `username`, `active`. \u26D4 Non-existent jobId throws error\n- `pg_cron_list_jobs`: List all jobs. Default `limit: 50` (use `0` for all). Optional `active` boolean filter. Returns `truncated` + `totalCount` when limited. Returns `hint` when jobs have no name\n- `pg_cron_job_run_details`: View execution history. Default `limit: 50`. Optional `jobId`, `status` ('running'|'succeeded'|'failed') filters. Returns `truncated` + `totalCount` when limited. Returns `summary` with counts\n- `pg_cron_cleanup_history`: Delete old run records. `olderThanDays`/`days` param (default: 7). Optional `jobId` to target specific job\n- `pg_cron_create_extension`: Enable pg_cron extension (idempotent). Requires superuser\n\n**Discovery**: `pg.cron.help()` returns `{methods, aliases, examples}` object\n\n## pgcrypto Tools\n\nCore: `createExtension()`, `hash()`, `hmac()`, `encrypt()`, `decrypt()`, `genRandomUuid()`, `genRandomBytes()`, `genSalt()`, `crypt()`\n\n- `pg_pgcrypto_create_extension`: Enable pgcrypto extension (idempotent). Returns `{success, message}`\n- `pg_pgcrypto_hash`: Hash data using digest algorithms. `algorithm`: 'md5', 'sha1', 'sha224', 'sha256', 'sha384', 'sha512'. `encoding`: 'hex' (default), 'base64'. Returns `{hash, algorithm, encoding, inputLength}`\n- `pg_pgcrypto_hmac`: HMAC authentication. Same algorithms as hash. Returns `{hmac, algorithm, encoding}`. `key` param for secret\n- `pg_pgcrypto_encrypt`: PGP symmetric encryption. `data` + `password`/`key` (aliases). Optional `options` for cipher config (e.g., 'cipher-algo=aes256'). Returns `{encrypted, encoding: 'base64'}`\n- `pg_pgcrypto_decrypt`: Decrypt PGP-encrypted data. `encryptedData`/`data` + `password`/`key` (aliases). Returns `{decrypted, verified}`. \u26D4 Throws on wrong key/corrupt data\n- `pg_pgcrypto_gen_random_uuid`: Generate UUID v4. Optional `count` (1-100, default 1). Returns `{uuid, uuids, count}` (`uuid` convenience property for single requests)\n- `pg_pgcrypto_gen_random_bytes`: Generate random bytes. `length` (1-1024). `encoding`: 'hex' (default), 'base64'. Returns `{randomBytes, length, encoding}`\n- `pg_pgcrypto_gen_salt`: Generate salt for crypt(). `type`: 'bf' (bcrypt, recommended), 'md5', 'xdes', 'des'. Optional `iterations` for bf (4-31) or xdes. Returns `{salt, type}`\n- `pg_pgcrypto_crypt`: Hash password with salt. Use stored hash as salt for verification. Returns `{hash, algorithm}`. Verification: `crypt(password, storedHash).hash === storedHash`\n\n**Password Workflow**: 1) `genSalt({type:'bf', iterations:10})` \u2192 2) `crypt({password, salt})` \u2192 store hash \u2192 3) Verify: `crypt({password, salt: storedHash})` and compare hashes\n\n**Top-Level Aliases**: `pg.pgcryptoHash()`, `pg.pgcryptoEncrypt()`, `pg.pgcryptoDecrypt()`, `pg.pgcryptoGenRandomUuid()`, etc.\n\n**Discovery**: `pg.pgcrypto.help()` returns `{methods, aliases, examples}` object\n\n## Code Mode Sandbox\n\nNo `setTimeout`, `setInterval`, `fetch`, or network access. Use `pg.core.readQuery()` for data access.\n\n\uD83D\uDCCA **Metrics Note**: `memoryUsedMb` measures heap delta (end - start). Negative values indicate memory freed during execution (e.g., GC ran).\n\n## Transactions\n\nCore: `begin()`, `commit()`, `rollback()`, `savepoint()`, `rollbackTo()`, `release()`, `execute()`\n\n**Transaction Lifecycle:**\n- `pg_transaction_begin`: Start new transaction. Returns `{transactionId, isolationLevel, message}`. Use `transactionId` for subsequent operations\n- `pg_transaction_commit`: Commit transaction, making all changes permanent. `transactionId`/`tx`/`txId` aliases\n- `pg_transaction_rollback`: Rollback transaction, discarding all changes. `transactionId`/`tx`/`txId` aliases\n\n**Savepoints:**\n- `pg_transaction_savepoint`: Create savepoint within transaction. `name`/`savepoint` + `transactionId`/`tx`/`txId`\n- `pg_transaction_rollback_to`: Rollback to savepoint, restoring database state to when the savepoint was created. \u26A0\uFE0F Undoes ALL work (data changes AND savepoints) created after the target savepoint\n- `pg_transaction_release`: Release savepoint, keeping all changes since it was created. `name`/`savepoint` aliases\n\n**Atomic Execution:**\n- `pg_transaction_execute`: Execute multiple statements atomically. Two modes:\n  - **Auto-commit**: Without `transactionId`\u2014auto-commits on success, auto-rollbacks on any error\n  - **Join existing**: With `transactionId`/`tx`/`txId`\u2014no auto-commit, caller controls via commit/rollback\n- `statements`: Array of `{sql: \"...\", params?: [...]}` objects. \u26A0\uFE0F Each object MUST have `sql` key\n- `isolationLevel`: Optional isolation level for new transactions ('READ COMMITTED', 'REPEATABLE READ', 'SERIALIZABLE')\n\n**Response Structures:**\n- `begin`: `{transactionId, isolationLevel: 'READ COMMITTED', message}`\n- `commit/rollback`: `{success, transactionId, message}`\n- `savepoint/release/rollbackTo`: `{success, transactionId, savepoint, message}`\n- `execute`: `{success, statementsExecuted, results: [{sql, rowsAffected, rowCount, rows?}], transactionId?}`\n\n**Discovery**: `pg.transactions.help()` returns `{methods, methodAliases, examples}`";
+export declare const SERVER_INSTRUCTIONS = "# postgres-mcp Code Mode\n\n## \u26A0\uFE0F Critical Gotchas\n\n1. **Transactions**: `pg.transactions.execute({statements: [{sql: \"...\"}]})` auto-commits on success, auto-rollbacks on error. To join existing transaction: `{transactionId: txId, statements: [...]}` (no auto-commit, caller controls)\n2. **pg_write_query**: \u26D4 Throws for SELECT\u2014use `pg_read_query` for SELECT statements\n3. **pg_upsert/pg_create_table**: `schema.table` format auto-parses (e.g., `'myschema.users'` \u2192 schema: 'myschema', table: 'users')\n4. **pg_create_table columns**: `notNull`, `defaultValue` (string literals auto-quoted; numbers/booleans auto-coerced; `now()` \u2192 `CURRENT_TIMESTAMP`), `check`, `references` (object or string `\"table(column)\"` syntax)\n5. **pg_create_table constraints**: `constraints` array only accepts `{type: 'unique'|'check'}`. Primary keys: use `column.primaryKey` or top-level `primaryKey: ['col1', 'col2']`\n6. **pg_create_index expression**: Columns can be expressions like `LOWER(name)` or `UPPER(email)`\u2014auto-detected. \u26A0\uFE0F Cast syntax (`::`) requires raw SQL via `pg_write_query`\n7. **pg_list_objects type**: Use `type` (singular string) or `types` (array). Auto-converts: `{type: 'table'}` \u2261 `{types: ['table']}`\n8. **pg_object_details**: Accepts: `name`, `objectName`, `object`, or `table`. Use `type`/`objectType` for type hint (supports: table, view, materialized_view, partitioned_table, function, sequence, index)\n9. **pg_exists optional WHERE**: `where`/`condition`/`filter` is optional. Without it, checks if table has any rows\n10. **pg_describe_table**: Returns columns, foreignKeys, primaryKey, indexes, constraints. For listing ALL database indexes (not table-specific), use `pg_get_indexes` without `table` param\n11. **pg_vector_insert updateExisting**: Uses direct UPDATE (avoids NOT NULL constraint issues vs INSERT mode)\n12. **pg_get_indexes without table**: Returns ALL database indexes (potentially large). Use `table` param for specific table\n13. **pg_upsert/pg_batch_insert RETURNING**: `returning` param must be array of column names: `[\"id\", \"name\"]`. \u26D4 `\"*\"` wildcard not supported\n14. **Small tables**: Optimizer correctly uses Seq Scan for <1000 rows\u2014this is expected behavior\n\n## \uD83D\uDD04 Response Structures\n\n| Tool | Returns | Notes |\n|------|---------|-------|\n| `pg_read_query` | `{rows, rowCount, fields?}` | `fields` contains column metadata (name, dataTypeID) |\n| `pg_write_query` | `{rowsAffected, affectedRows, rows?}` | `rows` only with RETURNING clause. DDL statements return `rowsAffected: 0`. \u26D4 Throws for SELECT |\n| `pg_upsert` | `{success, operation, rowsAffected, rowCount, rows?}` | `operation: 'insert'|'update'`. `rows` only with RETURNING clause |\n| `pg_batch_insert` | `{success, rowsAffected, affectedRows, insertedCount, rowCount, rows?}` | Empty objects use DEFAULT VALUES. \u26A0\uFE0F BIGINT > 2^53 loses precision |\n| `pg_create_table` | `{success, table, sql, compositePrimaryKey?}` | `table` = schema-qualified name. `compositePrimaryKey` only when composite PK used |\n| `pg_drop_table` | `{success, dropped, existed}` | `existed` indicates whether table was present before drop |\n| `pg_create_index` | `{success, index, indexName, table, sql, ifNotExists?, alreadyExists?, message?}` | `alreadyExists`/`message` only with `ifNotExists: true` when index pre-exists |\n| `pg_drop_index` | `{success, index, existed, sql}` | `existed` indicates whether index was present before drop |\n| `pg_truncate` | `{success, table, cascade, restartIdentity}` | `cascade`/`restartIdentity` reflect the options used |\n| `pg_count` | `{count: N}` | Use `params` for placeholders: `where: 'id=$1', params: [5]`. DISTINCT: use `pg_read_query` |\n| `pg_exists` | `{exists: bool, mode, hint?}` | `params` for placeholders. `mode: 'filtered'|'any_rows'` |\n| `pg_get_indexes` | `{indexes, count, totalCount?}` | Default `limit: 100` without `table`. Use `schema`/`limit` to filter. Index objects have `name`, `type`, `columns` |\n| `pg_list_objects` | `{objects, count, totalCount, byType}` | Use `limit` to cap results, `type`/`types` to filter |\n| `pg_object_details` | `{name, schema, type, returnType?, ...}` | Functions: `returnType` alias. Views/Mat. views: `definition`. Tables: equivalent to `pg_describe_table` (columns, primaryKey, indexes, constraints, foreignKeys) |\n| `pg_analyze_db_health` | `{cacheHitRatio, databaseSize, tableStats, unusedIndexes, tablesNeedingVacuum, connections, bloat, isReplica, overallScore, overallStatus}` | `cacheHitRatio`: `{ratio, heap, index, status}`. `overallStatus`: `healthy|needs_attention|critical`. Optional sections via `includeIndexes`, `includeVacuum`, `includeConnections` |\n| `pg_describe_table` | `{name, schema, type, owner, rowCount, columns, primaryKey, indexes, constraints, foreignKeys}` | Columns include `notNull` (alias for `!nullable`), `foreignKey`. `constraints` includes PK, UNIQUE, CHECK, NOT NULL. \u26A0\uFE0F `rowCount: -1` = stale/missing statistics (run ANALYZE on the table). Small tables (<~50 rows) may show -1 until first ANALYZE |\n| `pg_analyze_query_indexes` | `{plan, issues, recommendations}` | `verbosity`: 'summary' (default) or 'full'. Summary mode returns condensed plan |\n| `pg_list_tables` | `{tables, count, totalCount, truncated?, hint?}` | Use `schema` to filter, `limit` to cap results, `exclude` to hide extension schemas (e.g., `['cron', 'topology', 'partman']`) |\n| List operations | `{items, count}` | Access via `result.tables`, `result.views`, etc. |\n| `pg_jsonb_agg groupBy` | `{result: [{group_key, items}], count, grouped: true}` | Without groupBy: `{result: [...], count, grouped: false}` |\n| `pg_vector_aggregate` | `{average_vector, count}` or `{groups: [{group_key, average_vector, count}]}` | Without/with `groupBy` |\n| `pg_index_stats` | `{indexes, count, truncated?, totalCount?}` | Default 50 rows. Use `limit: 0` for all |\n| `pg_table_stats` | `{tables, count, truncated?, totalCount?}` | Default 50 rows. Use `limit: 0` for all |\n| `pg_vacuum_stats` | `{tables, count, truncated?, totalCount?}` | Default 50 rows. Use `limit: 0` for all |\n| `pg_stat_statements` | `{statements, count, truncated?, totalCount?}` | Default 20 rows. `orderBy` supported |\n| `pg_query_plan_stats` | `{queryPlanStats, count, truncated?, totalCount?}` | Default 20 rows. `truncateQuery: 0` for full text |\n| `pg_stat_activity` | `{connections, count}` | `includeIdle: true` to include idle connections |\n| `pg_locks` | `{locks}` | `showBlocked: true` switches to blocked/blocking pid format |\n| `pg_bloat_check` | `{tables, count}` | Tables with `live_tuples`, `dead_tuples`, `dead_pct` |\n| `pg_cache_hit_ratio` | `{heap_read, heap_hit, cache_hit_ratio}` | All fields nullable (0 tables = null). Flat response, differs from `pg_analyze_db_health.cacheHitRatio` |\n| `pg_seq_scan_tables` | `{tables, count, minScans, hint, truncated?, totalCount?}` | Default 50 rows. `minScans` default: 10 |\n| `pg_connection_pool_optimize` | `{current, config, waitEvents, recommendations}` | No params needed |\n| `pg_performance_baseline` | `{name, timestamp, metrics}` | `metrics`: `cache`, `tables`, `indexes`, `connections`, `databaseSize` |\n| `pg_duplicate_indexes` | `{duplicateIndexes, count, hint, truncated?, totalCount?}` | Default 50 rows. `duplicate_type`: EXACT_DUPLICATE, OVERLAPPING, SUBSET |\n| `pg_query_plan_compare` | `{query1, query2, analysis, fullPlans}` | `analysis.costDifference` + `recommendation` |\n| `pg_unused_indexes` | `{unusedIndexes, count, hint, truncated?, totalCount?}` | Default 20 rows. `summary: true` \u2192 `{summary, bySchema, totalCount}` |\n\n## API Mapping\n\n`pg_group_action` \u2192 `pg.group.action()` (group prefixes dropped: `pg_jsonb_extract` \u2192 `pg.jsonb.extract()`)\n\n**Top-Level Core Aliases**: All starter tools available directly: `pg.readQuery()`, `pg.writeQuery()`, `pg.listTables()`, `pg.describeTable()`, `pg.createTable()`, `pg.dropTable()`, `pg.count()`, `pg.exists()`, `pg.upsert()`, `pg.batchInsert()`, `pg.truncate()`, `pg.createIndex()`, `pg.dropIndex()`, `pg.getIndexes()`, `pg.listObjects()`, `pg.objectDetails()`, `pg.listExtensions()`, `pg.analyzeDbHealth()`, `pg.analyzeQueryIndexes()`, `pg.analyzeWorkloadIndexes()`\n\n**Positional args work**: `readQuery(\"SELECT...\")`, `exists(\"users\", \"id=1\")`, `createIndex(\"users\", [\"email\"])`\n\n**Discovery**: `pg.help()` returns `{group: methods[]}` mapping (e.g., `{core: ['readQuery', ...], jsonb: [...]}`). `pg.core.help()`, `pg.jsonb.help()` for group-specific methods.\n\n## Format Auto-Resolution\n\n- **Schema.Table**: `'public.users'` auto-parses to `{schema: 'public', table: 'users'}`\n- **JSONB Paths**: Both `'a.b.c'` (string) and `['a','b','c']` (array) work. Use array for literal dots: `[\"key.with.dots\"]`\n- **Aliases**: Common parameter variations resolve automatically (e.g., `query`/`sql`, `table`/`tableName`)\n\n---\n\n## Vector Tools\n\n\u26A0\uFE0F **Large Vectors**: Direct MCP tool calls may truncate vectors >256 dimensions due to JSON-RPC message size limits. For vectors \u2265256 dimensions (e.g., OpenAI 1536-dim, local 384-dim), use Code Mode: `await pg.vector.search({table, column, vector, limit})`\n\n- `pg_vector_search`: Supports `schema.table` format (auto-parsed). Returns `{results: [...], count, metric}`. Use `select: [\"id\", \"name\"]` to include identifying columns. Without select, only returns distance. `filter` = `where`. \u26A0\uFE0F Vectors read from DB are strings\u2014parse before passing: `vec.replace(/^\\[|\\]$/g, '').split(',').map(Number)`\n- `pg_vector_insert`: Supports `schema.table` format (auto-parsed). Use `updateExisting` + `conflictColumn` + `conflictValue` for UPDATE mode. `additionalColumns` is applied in both INSERT and UPDATE modes\n- `pg_vector_batch_insert`: `vectors` expects `[{vector: [...], data?: {...}}]` objects, not raw arrays\n- `pg_vector_normalize`: Returns `{normalized: [...], magnitude: N}`. Note: `magnitude` is the **original** vector length (not 1)\n- `pg_vector_aggregate`: Supports `schema.table` format (auto-parsed). \u26D4 Validates column is vector type. Returns `{average_vector: {preview, dimensions, truncated}, count}` or `{groups: [{group_key, average_vector, count}]}` with groupBy. \u26A0\uFE0F `groupBy` only supports simple column names (not expressions)\n- `pg_vector_dimension_reduce`: Direct mode returns `{reduced: [...], originalDimensions, targetDimensions}`. Table mode returns `{rows: [{id, original_dimensions, reduced}], processedCount, summarized}`. Default `summarize: true` in table mode returns compact `{preview, dimensions, truncated}` format. Use `summarize: false` for full vectors\n- `pg_vector_distance`: Calculate distance between two vectors. `metric`: 'l2' (default), 'cosine', 'inner_product'. Returns `{distance, metric}`\n- `pg_vector_cluster`: `clusters` = `k`. Returns centroids with `{preview, dimensions, truncated}` format for large vectors (>10 dims)\u2014use `pg_vector_distance` to assign rows\n- `pg_vector_create_index`: Use `type` (or alias `method`) with values 'ivfflat' or 'hnsw'. IVFFlat: `lists` param. HNSW: `m`, `efConstruction` params\n- `pg_vector_performance`: Auto-generates testVector from first row if omitted. Returns `testVectorSource: 'auto-generated from first row'|'user-provided'`\n- `pg_vector_validate`: Returns `{valid: bool, vectorDimensions}`. Empty vector `[]` returns `{valid: true, vectorDimensions: 0}`\n- \u26D4 `pg_vector_embed`: Demo only (hash-based). Use OpenAI/Cohere for production.\n- `pg_hybrid_search`: Supports `schema.table` format (auto-parsed). Combines vector similarity and full-text search with weighted scoring. `textColumn` auto-detects type: uses tsvector columns directly, wraps text columns with `to_tsvector()`. Code mode alias: `pg.hybridSearch()` \u2192 `pg.vector.hybridSearch()`\n- \uD83D\uDCDD **Error Handling**: Vector tools return `{success: false, error: \"...\", suggestion: \"...\"}` for validation/semantic errors (dimension mismatch, non-vector column, table not found). Check `success` field before processing results.\n\n## JSONB Tools\n\n- `pg_jsonb_extract`: Returns null if path doesn't exist\n- `pg_jsonb_insert`: Index -1 inserts BEFORE last element; use `insertAfter: true` to append. \u26A0\uFE0F Use array format `[-1]` not string `\"[-1]\"` for negative indices\n- `pg_jsonb_set`: `createMissing=true` creates full nested paths; initializes NULL columns to `{}`. Empty path (`''` or `[]`) replaces entire column value\n- `pg_jsonb_strip_nulls`: \u26A0\uFE0F Requires `where`/`filter` clause\u2014write operations must be targeted. Use `preview: true` to see changes first\n- `pg_jsonb_agg`: Supports AS aliases in select: `[\"id\", \"metadata->>'name' AS name\"]`. \u26A0\uFE0F `->>` returns text\u2014use `->` to preserve JSON types\n- `pg_jsonb_object`: Use `data`, `object`, or `pairs` parameter: `{data: {name: \"John\", age: 30}}`. Returns `{object: {...}}`\n- `pg_jsonb_normalize`: `flatten` doesn't descend into arrays; `keys` returns text (use `pairs` for JSON types)\n- `pg_jsonb_stats`: Returns column-level statistics. `topKeysLimit` controls key count (default: 20). \u26A0\uFE0F `typeDistribution` null type = SQL NULL columns (entire column NULL, not JSON `null` literal). Use `sqlNullCount` for explicit count\n- \u26D4 **Object-only tools**: `diff`, `merge`, `keys`, `indexSuggest`, `securityScan`, `stats`\u2014topKeys require JSONB objects, throw descriptive errors for arrays\n- \u26D4 **Array-only tools**: `insert`\u2014requires JSONB arrays, throws errors for objects\n- \uD83D\uDCDD `normalize` modes: `pairs`/`keys`/`flatten` for objects; `array` for arrays\n\n**Top-Level Aliases**: `pg.jsonbExtract()`, `pg.jsonbSet()`, `pg.jsonbInsert()`, `pg.jsonbDelete()`, `pg.jsonbContains()`, `pg.jsonbPathQuery()`, `pg.jsonbAgg()`, `pg.jsonbObject()`, `pg.jsonbArray()`, `pg.jsonbKeys()`, `pg.jsonbStripNulls()`, `pg.jsonbTypeof()`, `pg.jsonbValidatePath()`, `pg.jsonbMerge()`, `pg.jsonbNormalize()`, `pg.jsonbDiff()`, `pg.jsonbIndexSuggest()`, `pg.jsonbSecurityScan()`, `pg.jsonbStats()`\n\n## Stats Tools\n\n- All stats tools support `schema.table` format (auto-parsed, embedded schema takes priority over explicit `schema` param)\n- `timeSeries`: Both `timeColumn` (must be timestamp/date) and `valueColumn` (must be numeric) are validated upfront with clear error messages. Aliases: `time`\u2192`timeColumn`, `value`\u2192`valueColumn`. `interval` accepts: `second`, `minute`, `hour`, `day`, `week`, `month`, `year` (keywords, PostgreSQL format, or plurals). Default `limit: 100` time buckets. Use `limit: 0` for no limit. Returns `truncated` and `totalCount` indicators when default limit is applied. **groupBy payloads**: Default `groupLimit: 20` groups. Returns `truncated` + `totalGroupCount` when groups are limited. Use `groupLimit: 0` for all groups\n- `correlation`: Use `column1`/`column2` or aliases `x`/`y` for column names\n- `distribution`: Returns `skewness`, `kurtosis` (excess). `buckets` must be > 0. **groupBy payloads**: Default `groupLimit: 20` groups (prevents large payloads with many histogram buckets per group). Returns `truncated` + `totalGroupCount` when groups are limited. Use `groupLimit: 0` for all groups\n- `sampling`: Defaults to `random` method with 20 rows (optimized for LLM context). `sampleSize` always takes precedence over `percentage`. \u26A0\uFE0F `percentage` param only works with `bernoulli`/`system` methods\u2014ignored for default `random` method. Default limit of 100 rows applied to `bernoulli`/`system` with `percentage` to prevent large payloads. Returns `truncated` and `totalSampled` when TABLESAMPLE returns more rows than limit\n- `percentiles`: Accepts 0-1 or 0-100 (auto-normalized). \u26A0\uFE0F Use consistent scale\u2014mixing (e.g., `[0.1, 50]`) produces unexpected keys and returns a `warning` field explaining the issue. Empty array \u2192 defaults [0.25, 0.5, 0.75]\n- `hypothesis`: Returns nested `results` object containing `pValue` (two-tailed), `testStatistic`, `interpretation`, `sampleMean`, `sampleStdDev`. Access via `hyp.results.pValue`. Use `populationStdDev` for z-test, otherwise defaults to t-test\n- `regression`: Use `xColumn`/`yColumn`, aliases `x`/`y`, or `column1`/`column2` (for consistency with correlation). Returns nested `regression` object containing `slope`, `intercept`, `rSquared`, `equation`, `avgX`, `avgY`, `sampleSize`. Access via `reg.regression.slope`\n- \u26A0\uFE0F WARNING: `sampling` with `system` method unreliable for small tables\u2014use `bernoulli` or `random`\n\n**Top-Level Aliases**: `pg.descriptive()`, `pg.percentiles()`, `pg.correlation()`, `pg.regression()`, `pg.timeSeries()`, `pg.distribution()`, `pg.hypothesis()`, `pg.sampling()`\n\n## Text Tools\n\n- `pg_text_search`/`pg_text_rank`: Column must be `text` type\u2014pre-built `tsvector` columns are **not** supported (wrap with `to_tsvector()` fails on tsvector input). Use `pg_read_query` with raw FTS SQL for tsvector columns\n- `pg_create_fts_index`: Returns `{success, index, config, skipped}`. `skipped: true` = index already existed (IF NOT EXISTS). `ifNotExists` defaults to `true`\n\n## Performance Tools\n\nCore (20 methods): `explain()`, `explainAnalyze()`, `explainBuffers()`, `indexStats()`, `tableStats()`, `statStatements()`, `statActivity()`, `locks()`, `bloatCheck()`, `cacheHitRatio()`, `seqScanTables()`, `indexRecommendations()`, `queryPlanCompare()`, `baseline()`, `connectionPoolOptimize()`, `partitionStrategySuggest()`, `unusedIndexes()`, `duplicateIndexes()`, `vacuumStats()`, `queryPlanStats()`\n\nWrappers (3): `blockingQueries()`\u2192`locks({showBlocked:true})`, `longRunningQueries({ seconds | minDuration }?)` filters by duration (returns `{longRunningQueries, count, threshold}`), `analyzeTable({ table })` runs ANALYZE (accepts `schema.table` format)\n\n- `explain({ sql, format?, params? })`: Supports `format: 'text'|'json'|'yaml'|'xml'`. Default: text. Use `params: [value]` for `$1, $2` placeholders\n- `explainAnalyze({ sql, format?, params? })`: Same format/params options as explain\n- `explainBuffers({ sql, params? })`: Always returns JSON format (includes buffer statistics)\n- `indexRecommendations({ sql?, params? })`: Pass `params: [value]` for parameterized queries (e.g., `sql: 'SELECT * FROM orders WHERE id = $1', params: [5]`)\n- `queryPlanCompare({ query1, query2, params1?, params2? })`: Compare two query plans. Use `params1`/`params2` for parameterized queries\n- `partitionStrategySuggest({ table })`: Accepts `schema.table` format (auto-parsed) or separate `table` + `schema` params\n- \u26A0\uFE0F **Data Type Awareness**: Query literals must match column types exactly\u2014`WHERE sensor_id = 1` (integer), not `'sensor_1'` (string)\n\nAliases: `cacheStats`\u2192`cacheHitRatio`, `queryStats`\u2192`statStatements`, `activity`\u2192`statActivity`, `vacuum`\u2192`vacuumStats`, `indexUsage`\u2192`indexStats`, `bloatEstimate`/`bloat`\u2192`bloatCheck`, `runningQueries`\u2192`longRunningQueries`\n\n\uD83D\uDCE6 **AI-Optimized Payloads**: Tools return limited results by default to reduce context size:\n- `indexStats({ limit? })`: Default 50 rows. Returns `truncated: true` + `totalCount` when limited. Use `limit: 0` for all\n- `tableStats({ limit? })`: Default 50 rows. Returns `truncated: true` + `totalCount` when limited. Use `limit: 0` for all\n- `vacuumStats({ limit? })`: Default 50 rows. Same truncation indicators. Use `limit: 0` for all\n- `statStatements({ limit?, orderBy? })`: Default 20 rows. Returns `truncated: true` + `totalCount` when limited. Use `limit: 0` for all\n- `unusedIndexes({ limit?, summary? })`: Default 20 rows. Use `summary: true` for aggregated stats by schema\n- `queryPlanStats({ limit?, truncateQuery? })`: Default 20 rows, queries truncated to 100 chars. Use `truncateQuery: 0` for full text\n\n\uD83D\uDCCD **Code Mode Note**: `pg_performance_baseline` \u2192 `pg.performance.baseline()` (not `performanceBaseline`). `indexRecommendations` accepts `query` alias for `sql`\n\n**Top-Level Aliases**: `pg.explain()`, `pg.explainAnalyze()`, `pg.cacheHitRatio()`, `pg.indexStats()`, `pg.tableStats()`, `pg.indexRecommendations()`, `pg.bloatCheck()`, `pg.vacuumStats()`, `pg.unusedIndexes()`, `pg.duplicateIndexes()`, `pg.seqScanTables()`\n\n## Monitoring Tools\n\nCore: `databaseSize()`, `tableSizes()`, `connectionStats()`, `showSettings()`, `capacityPlanning()`, `uptime()`, `serverVersion()`, `recoveryStatus()`, `replicationStatus()`, `resourceUsageAnalyze()`, `alertThresholdSet()`\n\n- `databaseSize()`: Returns `{bytes: number, size: string}`. Optional `database` param for specific db\n- `tableSizes({ limit?, schema? })`: Default limit 50. Returns `{tables: [...], count, truncated?, totalCount?}`. `truncated: true` + `totalCount` when limited. Use `limit: 0` for all\n- `connectionStats()`: Returns `{byDatabaseAndState, totalConnections: number, maxConnections: number}`\n- `showSettings({ setting?, limit? })`: Default limit 50 when no pattern. Returns `{settings: [...], count, truncated?, totalCount?}`. Accepts `pattern`, `setting`, or `name`. Exact names auto-match; `%` for LIKE patterns\n- `capacityPlanning({days: 90})`: `days` = `projectionDays`. Returns `{current, growth, projection, recommendations}` with numeric fields. \u26D4 Negative days rejected\n- `uptime()`: Returns `{start_time: string, uptime: {days, hours, minutes, seconds, milliseconds}}`\n- `serverVersion()`: Returns `{full_version: string, version: string, version_num: number}`\n- `recoveryStatus()`: Returns `{in_recovery: boolean, last_replay_timestamp: string|null}`\n- `replicationStatus()`: Returns `{role: 'primary'|'replica', replicas: [...]}` for primary, or `{role: 'replica', replay_lag, ...}` for replica\n- `resourceUsageAnalyze()`: Returns `{backgroundWriter, checkpoints, connectionDistribution, bufferUsage, activity, analysis}` with all counts as numbers\n- `alertThresholdSet({metric?: 'connection_usage'})`: Returns recommended thresholds. Invalid metric returns `{success: false, error: \"...\"}`. Valid metrics: connection_usage, cache_hit_ratio, replication_lag, dead_tuples, long_running_queries, lock_wait_time\n\n\uD83D\uDCE6 **AI-Optimized Payloads**: Tools return limited results by default to reduce context size:\n- `tableSizes({ limit? })`: Default 50 rows. Returns `truncated: true` + `totalCount` when limited. Use `limit: 0` for all\n- `showSettings({ limit? })`: Default 50 rows when no pattern specified. Use `limit: 0` for all or specify a pattern\n\nAliases: `tables`\u2192`tableSizes`, `connections`\u2192`connectionStats`, `settings`/`config`\u2192`showSettings`, `alerts`/`thresholds`\u2192`alertThresholdSet`\n\n**Top-Level Aliases**: `pg.databaseSize()`, `pg.tableSizes()`, `pg.connectionStats()`, `pg.serverVersion()`, `pg.uptime()`, `pg.showSettings()`, `pg.recoveryStatus()`, `pg.replicationStatus()`, `pg.capacityPlanning()`, `pg.resourceUsageAnalyze()`, `pg.alertThresholdSet()`\n\n## Admin Tools\n\nCore: `vacuum()`, `vacuumAnalyze()`, `analyze()`, `reindex()`, `cluster()`, `setConfig()`, `reloadConf()`, `resetStats()`, `cancelBackend()`, `terminateBackend()`\n\n- All admin tools support `schema.table` format (auto-parsed, embedded schema takes priority over explicit `schema` param)\n- `vacuum({ table?, full?, analyze?, verbose? })`: Without `table`, vacuums ALL tables. `verbose` output goes to PostgreSQL server logs\n- `reindex({ target, name?, concurrently? })`: Targets: 'table', 'index', 'schema', 'database'. `database` target defaults to current db when `name` omitted\n- `cluster()`: Without args, re-clusters all previously-clustered tables. With args, requires BOTH `table` AND `index`\n- `setConfig({ name, value, isLocal? })`: `isLocal: true` applies only to current transaction\n- `cancelBackend({ pid })`: Graceful query cancellation\u2014returns `{success: false}` for invalid PID (no error thrown)\n- `terminateBackend({ pid })`: Forceful connection termination\u2014use with caution\n\nAliases: `tableName`\u2192`table`, `indexName`\u2192`index`, `param`/`setting`\u2192`name`, `processId`\u2192`pid`\n\n**Top-Level Aliases**: `pg.vacuum()`, `pg.vacuumAnalyze()`, `pg.analyze()`, `pg.reindex()`, `pg.cluster()`, `pg.setConfig()`, `pg.reloadConf()`, `pg.resetStats()`, `pg.cancelBackend()`, `pg.terminateBackend()`\n\n**Discovery**: `pg.admin.help()` returns `{methods, methodAliases, examples}` object\n\n**Response structures**:\n- `vacuum()` / `vacuumAnalyze()`: `{success, message, table?, schema?, hint?}` (hint present when verbose: true)\n- `analyze()`: `{success, message, table?, schema?, columns?}`\n- `reindex()`: `{success, message}`\n- `cluster()`: `{success, message, table?, index?}` (table/index present for table-specific cluster)\n- `setConfig()`: `{success, message, parameter, value}`\n- `reloadConf()` / `resetStats()`: `{success, message}`\n- `cancelBackend()` / `terminateBackend()`: `{success, message}`\n\n## Backup Tools\n\nCore: `dumpTable()`, `dumpSchema()`, `copyExport()`, `copyImport()`, `createBackupPlan()`, `restoreCommand()`, `physical()`, `restoreValidate()`, `scheduleOptimize()`\n\nResponse Structures:\n- `dumpTable`: `{ddl, type, note, insertStatements?}` \u2014 `insertStatements` only with `includeData: true` (separate field from `ddl`)\n- `copyExport`: `{data, rowCount, truncated?, limit?}` \u2014 `data` contains CSV/text content. `truncated: true` + `limit` when rows returned equals applied limit (indicating more rows likely exist)\n- `copyImport`: `{command, stdinCommand, notes}` \u2014 Both file and stdin COPY commands\n- `createBackupPlan`: `{strategy: {fullBackup, walArchiving}, estimates}`\n- `restoreCommand`: `{command, warnings?, notes}` \u2014 Warnings when `database` omitted\n- `restoreValidate`: `{note?, validationSteps: [{step, name, command?, commands?, note?}], recommendations}` \u2014 Top-level `note` when `backupType` omitted (defaults to pg_dump). Step-level `note` for non-command steps\n- `physical`: `{command, notes, requirements}`\n- `scheduleOptimize`: `{analysis, recommendation, commands}`\n\n\uD83D\uDCE6 **AI-Optimized Payloads**: `copyExport` limits results to 500 rows by default to prevent large payloads. Use `limit: 0` for all rows, or specify a custom limit.\n\n- `pg_copy_export`: Use `query`/`sql` OR `table`. \u26A0\uFE0F If both provided, `query` takes precedence with warning. Supports `schema.table` format (auto-parsed, takes priority over `schema` param). Format: `csv` (default, comma-delimited), `text` (tab-delimited). Both formats support `header: true` (default). \u26D4 `binary` not supported via MCP\u2014use `pg_dump_schema` for binary exports. Default `limit: 500` (use `0` for all rows). Optional `delimiter` to customize\n- `pg_dump_table`: Returns `ddl` + `insertStatements` when `includeData: true`. Supports sequences (`type: 'sequence'`), views (`type: 'view'`), and partitioned tables (`type: 'partitioned_table'` with `PARTITION BY` clause). **PRIMARY KEYS, INDEXES, CONSTRAINTS NOT included**\u2014use `pg_get_indexes`/`pg_get_constraints`. Supports `schema.table` format\n- `pg_dump_schema`: Generates pg_dump command. Optional `schema`, `table`, `filename`\n- `pg_copy_import`: Generates COPY FROM command. Supports `schema.table` format (auto-parsed, takes priority over `schema` param). `columns` array, `filePath`, `format`, `header`, `delimiter`\n- `pg_restore_command`: Include `database` parameter for complete command. Optional `schemaOnly`, `dataOnly`\n- `pg_create_backup_plan`: Generates backup strategy with cron schedule. `frequency`: 'hourly'|'daily'|'weekly', `retention` count\n- `pg_backup_physical`: Generates pg_basebackup command. `format`: 'plain'|'tar' (default: 'tar'), `checkpoint`: 'fast'|'spread', `compress`: 0-9\n- `pg_restore_validate`: Generates validation commands. `backupType`: 'pg_dump' (default)|'pg_basebackup'\n- `pg_backup_schedule_optimize`: Analyzes database activity patterns and recommends optimal backup schedule\n\n**Top-Level Aliases**: `pg.dumpTable()`, `pg.dumpSchema()`, `pg.copyExport()`, `pg.copyImport()`, `pg.createBackupPlan()`, `pg.restoreCommand()`, `pg.restoreValidate()`, `pg.physical()`, `pg.backupPhysical()`, `pg.scheduleOptimize()`, `pg.backupScheduleOptimize()`\n\n## Text Tools\n\nDefaults: `threshold`=0.3 (use 0.1-0.2 for partial), `maxDistance`=3 (use 5+ for longer strings)\n\n- All text tools support `schema.table` format (auto-parsed, embedded schema takes priority over explicit `schema` param)\n- `pg_text_search`: Supports both `column` (singular string) and `columns` (array). Either is valid\u2014`column` auto-converts to array\n- `pg_trigram_similarity`, `pg_fuzzy_match`, `pg_regexp_match`, `pg_like_search`: All default to 100 results to prevent large payloads. Use `limit: 0` for all rows\n- `pg_fuzzy_match`: Levenshtein returns distance (lower=better). Soundex/metaphone return phonetic codes (exact match only). \u26D4 Invalid `method` values throw error with valid options\n- `pg_text_normalize`: Removes accents only (unaccent). Does NOT lowercase/trim\n- \uD83D\uDCCD **Table vs Standalone**: `normalize`, `sentiment`, `toVector`, `toQuery`, `searchConfig` are standalone (text input only). For phonetic matching: use `pg_fuzzy_match` with `method: 'soundex'|'metaphone'` (direct MCP), or `pg.text.soundex()`/`pg.text.metaphone()` (Code Mode convenience wrappers that call fuzzyMatch internally)\n\n**Top-Level Aliases**: `pg.textSearch()`, `pg.textRank()`, `pg.textHeadline()`, `pg.textNormalize()`, `pg.textSentiment()`, `pg.textToVector()`, `pg.textToQuery()`, `pg.textSearchConfig()`, `pg.textTrigramSimilarity()`, `pg.textFuzzyMatch()`, `pg.textLikeSearch()`, `pg.textRegexpMatch()`, `pg.textCreateFtsIndex()`\n\n## Schema Tools\n\nCore: `listSchemas()`, `createSchema()`, `dropSchema()`, `listViews()`, `createView()`, `dropView()`, `listSequences()`, `createSequence()`, `dropSequence()`, `listFunctions()`, `listTriggers()`, `listConstraints()`\n\nResponse Structures:\n- `listSchemas()`: `{schemas: string[], count}`\n- `listViews({ includeMaterialized?, truncateDefinition?, limit? })`: `{views: [{schema, name, type, definition, definitionTruncated?}], count, hasMatViews, truncatedDefinitions?, truncated, note?}`. Default `limit: 50` (use `0` for all). Default `truncateDefinition: 500` chars (use `0` for full definitions). `truncated` always included (`true`/`false`)\n- `listSequences({ schema? })`: `{sequences: [{schema, name, owned_by}], count}`. Note: `owned_by` omits `public.` prefix for sequences in public schema (e.g., `users.id` not `public.users.id`)\n- `listFunctions({ schema?, limit?, exclude? })`: `{functions: [{schema, name, arguments, returns, language, volatility}], count, limit}`\n- `listTriggers({ schema?, table? })`: `{triggers: [{schema, table_name, name, timing, events, function_name, enabled}], count}`\n- `listConstraints({ schema?, table?, type? })`: `{constraints: [{schema, table_name, name, type, definition}], count}`. Type codes: `p`=primary_key, `f`=foreign_key, `u`=unique, `c`=check\n- `dropSchema/dropView/dropSequence`: All return `{existed: true/false}` to indicate if object existed before drop\n- `createSchema/createSequence` (with `ifNotExists`) and `createView` (with `orReplace`): Return `{alreadyExisted: true/false}` when the flag is set. Without `ifNotExists`/`orReplace`, the field is omitted\n\n- `pg_create_view`: Supports `schema.name` format (auto-parsed). Use `orReplace: true` for CREATE OR REPLACE. `checkOption`: 'cascaded', 'local', 'none'. \u26D4 OR REPLACE can add new columns but cannot rename/remove existing ones\u2014PostgreSQL limitation\n- `pg_create_sequence`: Supports `schema.name` format. Parameters: `start`, `increment`, `minValue`, `maxValue`, `cache`, `cycle`, `ownedBy`, `ifNotExists`\n- `pg_list_functions`: Default limit=500. Use `schema: 'public'`, `limit: 2000`, or `exclude: ['postgis', 'pg_trgm', 'ltree', 'citext', 'fuzzystrmatch', 'pg_stat_statements', 'hypopg', 'unaccent', 'pg_stat_kcache', 'pgcrypto', 'partman', 'vector', 'topology']` to filter. \u26A0\uFE0F `exclude` filters by **schema name** AND extension-owned functions. The `language` filter does NOT exclude extension functions\u2014use `exclude` alongside `language` for clean results. Note: Aggressive `exclude` may return 0 results if all functions belong to excluded extensions\n\n**Discovery**: `pg.schema.help()` returns `{methods, methodAliases, examples}` object\n\n## Partitioning Tools\n\n- `pg_create_partitioned_table`: `partitionBy` case-insensitive. Supports `schema.table` format for `name` (auto-parsed). `primaryKey` accepts array (e.g., `['id', 'event_date']`). \u26D4 `primaryKey`/`unique` must include partition key\u2014throws validation error otherwise\n- `pg_create_partition`: Use `parent`/`table`/`parentTable`. `forValues` is a raw SQL string: `\"FROM ('2024-01-01') TO ('2024-07-01')\"`, `\"IN ('US', 'CA')\"`, `\"WITH (MODULUS 4, REMAINDER 0)\"`. For DEFAULT partition, use `isDefault: true`. Supports `schema.table` format for `parent` (auto-parsed)\n- `pg_attach_partition`/`pg_detach_partition`: Support `schema.table` format for `parent` and `partition` (auto-parsed). For DEFAULT partition, use `isDefault: true` or `forValues: \"DEFAULT\"`\n- `pg_list_partitions`: Default `limit: 50` (use `0` for all). Returns `{partitions, count, truncated, totalCount?}`. Uses `bounds` field (consistent with `pg_partition_info`)\n- `pg_partition_info`: Returns `{tableInfo, partitions, totalSizeBytes}`. Uses `bounds` field\n- Both list/info tools support `schema.table` format (auto-parsed) and accept `table`, `parent`, `parentTable`, or `name` aliases\n- Response structures: `pg_create_partitioned_table` \u2192 `{success, table, partitionBy, partitionKey, primaryKey?}`. `pg_create_partition` \u2192 `{success, partition, parent, bounds, subpartitionBy?, subpartitionKey?}`. `pg_attach_partition` \u2192 `{success, parent, partition, bounds}`. `pg_detach_partition` \u2192 `{success, parent, partition}`\n- \u26A0\uFE0F Sub-partitioning: `subpartitionBy`/`subpartitionKey` on `pg_create_partition` makes a partition itself partitionable. The parent's `primaryKey` must include the sub-partition key column (PostgreSQL constraint)\n- \uD83D\uDCCD Code Mode: `pg.partitioning.create()` = `createPartition`, NOT `createPartitionedTable`\n\n## pg_partman Tools\n\n- `pg_partman_create_parent`: Interval uses PostgreSQL syntax ('1 day', '1 month') NOT keywords ('daily'). `startPartition` accepts 'now' shorthand for current date. Required params: `parentTable`/`table`, `controlColumn`/`control`/`column`, `interval`\n- `pg_partman_run_maintenance`: Without `parentTable`, maintains ALL partition sets. Returns `partial: true` when some tables are skipped. `orphaned` object groups orphaned configs with `count`, `tables`, and cleanup `hint`. `errors` array for other failures\n- `pg_partman_show_config`: Default `limit: 50` (use `0` for all). Returns `truncated` + `totalCount` when limited. `orphaned` flag per config. Supports `schema.table` or plain table name (auto-prefixes `public.`)\n- `pg_partman_show_partitions`: Default `limit: 50` (use `0` for all). Returns `truncated` + `totalCount` when limited. `parentTable` required. Supports `schema.table` format (auto-parsed)\n- `pg_partman_check_default`/`partition_data`: `parentTable` required. Supports `schema.table` format (auto-parsed)\n- `pg_partman_set_retention`: \u26A0\uFE0F **CAUTION: Default is DROP** \u2014 `retentionKeepTable: false` (default) = DROP partitions, `true` = detach only (safer). Pass `retention: null` to disable retention\n- `pg_partman_undo_partition`: `targetTable` MUST exist before calling. Requires both `parentTable` and `targetTable`/`target`. \u26A0\uFE0F Parent table and child partitions remain after undo\u2014use `DROP TABLE parent CASCADE` to clean up\n- `pg_partman_analyze_partition_health`: Default `limit: 50` (use `0` for all). Returns `truncated` + `totalCount` when limited. `summary.overallHealth`: 'healthy'|'warnings'|'issues_found'\n- \uD83D\uDCDD **Schema Resolution**: All partman tools auto-prefix `public.` when no schema specified in `parentTable`\n- \uD83D\uDCDD **Aliases**: `parentTable` accepts `table`, `parent`, `name`. `controlColumn` accepts `control`, `column`. `targetTable` accepts `target`\n\n## pg_stat_kcache Tools\n\nCore: `createExtension()`, `queryStats()`, `topCpu()`, `topIo()`, `databaseStats()`, `resourceAnalysis()`, `reset()`\n\n- `pg_kcache_query_stats`: Default `limit: 20` (use `0` for all). Returns `truncated` + `totalCount` when limited. `orderBy`: 'total_time' (default), 'cpu_time', 'reads', 'writes'. `queryPreviewLength`: chars for query preview (default: 100, max: 500, 0 for full). \u26D4 'calls' NOT valid for orderBy\u2014use `minCalls` param\n- `pg_kcache_resource_analysis`: Default `limit: 20` (use `0` for all). Returns `truncated` + `totalCount` when limited. `minCalls`, `queryPreviewLength` supported. Classifies queries as 'CPU-bound', 'I/O-bound', or 'Balanced'\n- `pg_kcache_top_cpu`: Top CPU-consuming queries. `limit` param (default: 10). `queryPreviewLength`: chars for query preview (default: 100, max: 500, 0 for full). Returns `truncated` + `totalCount` when limited\n- `pg_kcache_top_io`: `type`/`ioType` (alias): 'reads', 'writes', 'both' (default). `limit` param (default: 10). `queryPreviewLength`: chars for query preview (default: 100, max: 500, 0 for full). Returns `truncated` + `totalCount` when limited\n- `pg_kcache_database_stats`: Aggregated CPU/IO stats per database. Optional `database` param to filter specific db\n- `pg_kcache_reset`: Resets pg_stat_kcache AND pg_stat_statements statistics\n\n## citext Tools\n\nCore: `createExtension()`, `convertColumn()`, `listColumns()`, `analyzeCandidates()`, `compare()`, `schemaAdvisor()`\n\n- `pg_citext_create_extension`: Enable citext extension (idempotent). Returns `{success, message, usage}`\n- `pg_citext_convert_column`: Supports `schema.table` format (auto-parsed). \u26D4 Only allows text-based columns (text, varchar, character varying)\u2014non-text columns return `{success: false, error, allowedTypes, suggestion}`. When views depend on column, returns `{success: false, dependentViews, hint}`\u2014drop/recreate views manually. `col` alias for `column`. Returns `{previousType}` showing original type\n- `pg_citext_list_columns`: Default `limit: 100` (use `0` for all). Returns `{columns: [{table_schema, table_name, column_name, is_nullable, column_default}], count, totalCount, truncated}`. Optional `schema`, `limit` filters\n- `pg_citext_analyze_candidates`: Default `limit: 50` (use `0` for all). Default `excludeSystemSchemas: true` filters out extension schemas (cron, topology, partman, tiger) when no `schema`/`table` filter specified\u2014use `excludeSystemSchemas: false` to include all. Returns `truncated: true` + `totalCount` when results are limited. Scans tables for TEXT/VARCHAR columns matching common patterns (email, username, name, etc.). Optional `schema`, `table`, `limit`, `excludeSystemSchemas`, `patterns` filters. Returns `{candidates, count, totalCount, truncated, summary: {highConfidence, mediumConfidence}, recommendation, patternsUsed, excludedSchemas?}`\n- `pg_citext_compare`: Test case-insensitive comparison. Returns `{value1, value2, citextEqual, textEqual, lowerEqual, extensionInstalled}`\n- `pg_citext_schema_advisor`: Supports `schema.table` format (auto-parsed). Analyzes specific table. Returns `{table, recommendations: [{column, currentType, previousType?, recommendation, confidence, reason}], summary, nextSteps}`. `tableName` alias for `table`. Already-citext columns include `previousType: \"text or varchar (converted)\"`\n\n**Discovery**: `pg.citext.help()` returns `{methods, methodAliases, examples}` object\n\n## ltree Tools\n\nCore: `createExtension()`, `query()`, `match()`, `subpath()`, `lca()`, `listColumns()`, `convertColumn()`, `createIndex()`\n\n- `pg_ltree_create_extension`: Enable ltree extension (idempotent). Returns `{success, message}`\n- `pg_ltree_query`: Query hierarchical relationships. Supports `schema.table` format (auto-parsed). `mode`/`type`: 'ancestors', 'descendants' (default), 'exact'. Returns `{results, count, path, mode, isPattern}`. \u26A0\uFE0F Validates column is ltree type\u2014returns clear error for non-ltree columns\n- `pg_ltree_match`: Match paths using lquery pattern syntax (`*`, `*{1,2}`, `*.label.*`). Supports `schema.table` format. `pattern`/`lquery`/`query` aliases. Returns `{results, count, pattern}`\n- `pg_ltree_subpath`: Extract portion of ltree path. `offset`/`start`/`from` and `length`/`len` aliases. Negative `offset` counts from end. \u26A0\uFE0F Returns `{success: false, error, pathDepth}` for invalid offset (validated before PostgreSQL call)\n- `pg_ltree_lca`: Find longest common ancestor of multiple paths. Requires `paths` array (min 2). Returns `{longestCommonAncestor, hasCommonAncestor: bool, paths}`\n- `pg_ltree_list_columns`: List all ltree columns in database. Optional `schema` filter. Returns `{columns: [{table_schema, table_name, column_name, is_nullable, column_default}], count}`\n- `pg_ltree_convert_column`: Convert TEXT column to ltree. Supports `schema.table` format. `col` alias for `column`. Returns `{previousType}`. \u26A0\uFE0F When views depend on column, returns `{success: false, dependentViews, hint}`\u2014drop/recreate views manually\n- `pg_ltree_create_index`: Create GiST index on ltree column. Supports `schema.table` format. Auto-generates index name if `indexName` omitted. Returns `{indexName, indexType: 'gist', alreadyExists?}`\n\n**Discovery**: `pg.ltree.help()` returns `{methods, methodAliases, examples}` object. Top-level aliases available: `pg.ltreeQuery()`, `pg.ltreeMatch()`, etc.\n\n## PostGIS Tools\n\n**Geometry Creation:**\n- `pg_geocode`: Create point geometry from lat/lng. Returns `{geojson, wkt}`. \u26A0\uFE0F Validates bounds: lat \u00B190\u00B0, lng \u00B1180\u00B0\n- `pg_geometry_column`: Add geometry column to table. `ifNotExists` returns `{alreadyExists: true}`\n- `pg_spatial_index`: Create GiST spatial index. Auto-generates name if not provided. `ifNotExists` supported\n\n**Spatial Queries:**\n- `pg_distance`: Find geometries within distance from point. Returns `{results, count}` with `distance_meters`. \u26A0\uFE0F Validates point bounds\n- `pg_bounding_box`: Find geometries within lat/lng bounding box. Use `select` array for specific columns\n- `pg_intersection`: Find geometries intersecting a WKT/GeoJSON geometry. Auto-detects SRID from column\n- `pg_point_in_polygon`: Check if point is within table polygons. Returns `{containingPolygons, count}`. \u26A0\uFE0F Validates point bounds\n\n**Geometry Operations (Table-based):**\n- `pg_buffer`: Create buffer zone around table geometries. Default limit: 50 rows. Default simplify: 10m (set `simplify: 0` to disable). Returns `truncated: true` + `totalCount` when results are truncated. Use `limit: 0` for all rows\n- `pg_geo_transform`: Transform table geometries between SRIDs. Default limit: 50 rows. Returns `truncated: true` + `totalCount` when results are truncated. Use `limit: 0` for all rows. Auto-detects `fromSrid` from column metadata if not provided (returns `autoDetectedSrid: true`). `fromSrid`/`sourceSrid` and `toSrid`/`targetSrid` aliases\n- `pg_geo_cluster`: Spatial clustering (DBSCAN/K-Means). K-Means: If `numClusters` exceeds row count, automatically clamps to available rows with `warning` field. DBSCAN: Returns contextual `hints` array explaining parameter effects (e.g., \"All points formed single cluster\u2014decrease eps\") and `parameterGuide` explaining eps/minPoints trade-offs\n\n**Geometry Operations (Standalone WKT/GeoJSON):**\n- `pg_geometry_buffer`: Create buffer around WKT/GeoJSON. Returns `{buffer_geojson, buffer_wkt, distance_meters}`. Optional `simplify` param (meters) reduces polygon complexity\u2014returns `simplified`, `simplifyTolerance` when applied. \u26A0\uFE0F Returns `warning` if simplify tolerance is too high and geometry collapses to null\n- `pg_geometry_transform`: Transform WKT/GeoJSON between SRIDs. Returns `{transformed_geojson, transformed_wkt, fromSrid, toSrid}`\n- `pg_geometry_intersection`: Compute intersection of two geometries. Returns `{intersects, intersection_geojson, intersection_area_sqm}`. Normalizes SRID (4326) automatically\u2014safe to mix GeoJSON and WKT\n\n**Administration:**\n- `pg_postgis_create_extension`: Enable PostGIS extension (idempotent)\n- `pg_geo_index_optimize`: Analyze spatial indexes. Without `table` param, analyzes all spatial indexes\n\n**Code Mode Aliases:** `pg.postgis.addColumn()` \u2192 `geometryColumn`, `pg.postgis.indexOptimize()` \u2192 `geoIndexOptimize`, `pg.postgis.geoCluster()` \u2192 `pg_geo_cluster`, `pg.postgis.geoTransform()` \u2192 `pg_geo_transform`. Note: `pg.{group}.help()` returns `{methods, methodAliases, examples}`\n\n## Cron Tools (pg_cron)\n\nCore: `createExtension()`, `schedule()`, `scheduleInDatabase()`, `unschedule()`, `alterJob()`, `listJobs()`, `jobRunDetails()`, `cleanupHistory()`\n\n- `pg_cron_schedule`: Schedule a cron job. `schedule` supports standard cron (`0 5 * * *`) or interval (`1 second` to `59 seconds`). \u26A0\uFE0F Interval syntax only works for 1-59 seconds\u2014for 60+ seconds, use cron syntax (e.g., `* * * * *` for every minute). Use `name`/`jobName` for identification. `command`/`sql`/`query` aliases supported. Note: pg_cron allows duplicate job names; use unique names to avoid confusion when unscheduling\n- `pg_cron_schedule_in_database`: Schedule job in specific database. `database`/`db` aliases. Optional `username`, `active` params\n- `pg_cron_unschedule`: Remove job by `jobId` or `jobName`. If both provided, `jobName` takes precedence (with warning)\n- `pg_cron_alter_job`: Modify existing job. Can change `schedule`, `command`, `database`, `username`, `active`. \u26D4 Non-existent jobId returns error\n- `pg_cron_list_jobs`: List all jobs. Default `limit: 50` (use `0` for all). Optional `active` boolean filter. Returns `truncated` + `totalCount` when limited. Returns `hint` when jobs have no name\n- `pg_cron_job_run_details`: View execution history. Default `limit: 50`. Optional `jobId`, `status` ('running'|'succeeded'|'failed') filters. Returns `truncated` + `totalCount` when limited. Returns `summary` with counts\n- `pg_cron_cleanup_history`: Delete old run records. `olderThanDays`/`days` param (default: 7). Optional `jobId` to target specific job\n- `pg_cron_create_extension`: Enable pg_cron extension (idempotent). Requires superuser\n\n**Discovery**: `pg.cron.help()` returns `{methods, methodAliases, examples}` object\n\n## pgcrypto Tools\n\nCore: `createExtension()`, `hash()`, `hmac()`, `encrypt()`, `decrypt()`, `genRandomUuid()`, `genRandomBytes()`, `genSalt()`, `crypt()`\n\n- `pg_pgcrypto_create_extension`: Enable pgcrypto extension (idempotent). Returns `{success, message}`\n- `pg_pgcrypto_hash`: Hash data using digest algorithms. `algorithm`: 'md5', 'sha1', 'sha224', 'sha256', 'sha384', 'sha512'. `encoding`: 'hex' (default), 'base64'. Returns `{hash, algorithm, encoding, inputLength}`\n- `pg_pgcrypto_hmac`: HMAC authentication. Same algorithms as hash. Returns `{hmac, algorithm, encoding}`. `key` param for secret\n- `pg_pgcrypto_encrypt`: PGP symmetric encryption. `data` + `password`/`key` (aliases). Optional `options` for cipher config (e.g., 'cipher-algo=aes256'). Returns `{encrypted, encoding: 'base64'}`\n- `pg_pgcrypto_decrypt`: Decrypt PGP-encrypted data. `encryptedData`/`data` + `password`/`key` (aliases). Returns `{decrypted, verified}`. \u26D4 Throws on wrong key/corrupt data\n- `pg_pgcrypto_gen_random_uuid`: Generate UUID v4. Optional `count` (1-100, default 1). Returns `{uuid, uuids, count}` (`uuid` convenience property for single requests)\n- `pg_pgcrypto_gen_random_bytes`: Generate random bytes. `length` (1-1024). `encoding`: 'hex' (default), 'base64'. Returns `{randomBytes, length, encoding}`\n- `pg_pgcrypto_gen_salt`: Generate salt for crypt(). `type`: 'bf' (bcrypt, recommended), 'md5', 'xdes', 'des'. Optional `iterations` for bf (4-31) or xdes. Returns `{salt, type}`\n- `pg_pgcrypto_crypt`: Hash password with salt. Use stored hash as salt for verification. Returns `{hash, algorithm}`. Verification: `crypt(password, storedHash).hash === storedHash`\n\n**Password Workflow**: 1) `genSalt({type:'bf', iterations:10})` \u2192 2) `crypt({password, salt})` \u2192 store hash \u2192 3) Verify: `crypt({password, salt: storedHash})` and compare hashes\n\n**Top-Level Aliases**: `pg.pgcryptoHash()`, `pg.pgcryptoEncrypt()`, `pg.pgcryptoDecrypt()`, `pg.pgcryptoGenRandomUuid()`, etc.\n\n**Discovery**: `pg.pgcrypto.help()` returns `{methods, methodAliases, examples}` object\n\n## Code Mode Sandbox\n\nNo `setTimeout`, `setInterval`, `fetch`, or network access. Use `pg.core.readQuery()` for data access.\n\n\uD83D\uDCCA **Metrics Note**: `memoryUsedMb` measures heap delta (end - start). Negative values indicate memory freed during execution (e.g., GC ran).\n\n## Transactions\n\nCore: `begin()`, `commit()`, `rollback()`, `savepoint()`, `rollbackTo()`, `release()`, `execute()`\n\n**Transaction Lifecycle:**\n- `pg_transaction_begin`: Start new transaction. Returns `{transactionId, isolationLevel, message}`. Use `transactionId` for subsequent operations\n- `pg_transaction_commit`: Commit transaction, making all changes permanent. `transactionId`/`tx`/`txId` aliases\n- `pg_transaction_rollback`: Rollback transaction, discarding all changes. `transactionId`/`tx`/`txId` aliases\n\n**Savepoints:**\n- `pg_transaction_savepoint`: Create savepoint within transaction. `name`/`savepoint` + `transactionId`/`tx`/`txId`\n- `pg_transaction_rollback_to`: Rollback to savepoint, restoring database state to when the savepoint was created. \u26A0\uFE0F Undoes ALL work (data changes AND savepoints) created after the target savepoint\n- `pg_transaction_release`: Release savepoint, keeping all changes since it was created. `name`/`savepoint` aliases\n\n**Atomic Execution:**\n- `pg_transaction_execute`: Execute multiple statements atomically. Two modes:\n  - **Auto-commit**: Without `transactionId`\u2014auto-commits on success, auto-rollbacks on any error\n  - **Join existing**: With `transactionId`/`tx`/`txId`\u2014no auto-commit, caller controls via commit/rollback\n- `statements`: Array of `{sql: \"...\", params?: [...]}` objects. \u26A0\uFE0F Each object MUST have `sql` key\n- `isolationLevel`: Optional isolation level for new transactions ('READ COMMITTED', 'REPEATABLE READ', 'SERIALIZABLE')\n- Supports SELECT statements inside `statements`\u2014results include `rows` in the response for mixed read/write workflows\n\n**Aborted Transaction State:**\n- \u26A0\uFE0F If any statement in a transaction fails, PostgreSQL puts the transaction into an **aborted state**\n- In aborted state, only `ROLLBACK` or `ROLLBACK TO SAVEPOINT` commands are accepted\u2014all other commands will error\n- Use `pg_transaction_rollback` to end the transaction, or `pg_transaction_rollback_to` to recover to a savepoint\n- `pg_transaction_commit` on an aborted transaction will detect the state and report it (not silently rollback)\n\n**Response Structures:**\n- `begin`: `{transactionId, isolationLevel: 'READ COMMITTED', message}`\n- `commit/rollback`: `{success, transactionId, message}`\n- `savepoint/release/rollbackTo`: `{success, transactionId, savepoint, message}`\n- `execute`: `{success, statementsExecuted, results: [{sql, rowsAffected, rowCount, rows?}], transactionId?}`\n\n**Discovery**: `pg.transactions.help()` returns `{methods, methodAliases, examples}`";
 //# sourceMappingURL=ServerInstructions.d.ts.map

package/dist/constants/ServerInstructions.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"ServerInstructions.d.ts","sourceRoot":"","sources":["../../src/constants/ServerInstructions.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AACH,eAAO,MAAM,mBAAmB,sn4CA8YyD,CAAC"}
+{"version":3,"file":"ServerInstructions.d.ts","sourceRoot":"","sources":["../../src/constants/ServerInstructions.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AACH,eAAO,MAAM,mBAAmB,+llDAwbyD,CAAC"}

package/dist/constants/ServerInstructions.js

@@ -22,7 +22,7 @@ export const SERVER_INSTRUCTIONS = `# postgres-mcp Code Mode
 7. **pg_list_objects type**: Use \`type\` (singular string) or \`types\` (array). Auto-converts: \`{type: 'table'}\` ≡ \`{types: ['table']}\`
 8. **pg_object_details**: Accepts: \`name\`, \`objectName\`, \`object\`, or \`table\`. Use \`type\`/\`objectType\` for type hint (supports: table, view, materialized_view, partitioned_table, function, sequence, index)
 9. **pg_exists optional WHERE**: \`where\`/\`condition\`/\`filter\` is optional. Without it, checks if table has any rows
-10. **pg_describe_table**: Returns columns, foreignKeys, primaryKey—use \`pg_get_indexes\` separately for index details
+10. **pg_describe_table**: Returns columns, foreignKeys, primaryKey, indexes, constraints. For listing ALL database indexes (not table-specific), use \`pg_get_indexes\` without \`table\` param
 11. **pg_vector_insert updateExisting**: Uses direct UPDATE (avoids NOT NULL constraint issues vs INSERT mode)
 12. **pg_get_indexes without table**: Returns ALL database indexes (potentially large). Use \`table\` param for specific table
 13. **pg_upsert/pg_batch_insert RETURNING**: \`returning\` param must be array of column names: \`["id", "name"]\`. ⛔ \`"*"\` wildcard not supported
@@ -33,21 +33,41 @@ export const SERVER_INSTRUCTIONS = `# postgres-mcp Code Mode
 | Tool | Returns | Notes |
 |------|---------|-------|
 | \`pg_read_query\` | \`{rows, rowCount, fields?}\` | \`fields\` contains column metadata (name, dataTypeID) |
-| \`pg_write_query\` | \`{rowsAffected, affectedRows, rows?}\` | \`rows\` only with RETURNING clause. ⛔ Throws for SELECT |
-| \`pg_upsert\` | \`{operation, rowsAffected, rowCount, rows?}\` | \`operation: 'insert'|'update'\`. \`rows\` only with RETURNING clause |
-| \`pg_batch_insert\` | \`{rowsAffected, affectedRows, insertedCount, rows?}\` | Empty objects use DEFAULT VALUES. ⚠️ BIGINT > 2^53 loses precision |
+| \`pg_write_query\` | \`{rowsAffected, affectedRows, rows?}\` | \`rows\` only with RETURNING clause. DDL statements return \`rowsAffected: 0\`. ⛔ Throws for SELECT |
+| \`pg_upsert\` | \`{success, operation, rowsAffected, rowCount, rows?}\` | \`operation: 'insert'|'update'\`. \`rows\` only with RETURNING clause |
+| \`pg_batch_insert\` | \`{success, rowsAffected, affectedRows, insertedCount, rowCount, rows?}\` | Empty objects use DEFAULT VALUES. ⚠️ BIGINT > 2^53 loses precision |
+| \`pg_create_table\` | \`{success, table, sql, compositePrimaryKey?}\` | \`table\` = schema-qualified name. \`compositePrimaryKey\` only when composite PK used |
+| \`pg_drop_table\` | \`{success, dropped, existed}\` | \`existed\` indicates whether table was present before drop |
+| \`pg_create_index\` | \`{success, index, indexName, table, sql, ifNotExists?, alreadyExists?, message?}\` | \`alreadyExists\`/\`message\` only with \`ifNotExists: true\` when index pre-exists |
+| \`pg_drop_index\` | \`{success, index, existed, sql}\` | \`existed\` indicates whether index was present before drop |
+| \`pg_truncate\` | \`{success, table, cascade, restartIdentity}\` | \`cascade\`/\`restartIdentity\` reflect the options used |
 | \`pg_count\` | \`{count: N}\` | Use \`params\` for placeholders: \`where: 'id=$1', params: [5]\`. DISTINCT: use \`pg_read_query\` |
 | \`pg_exists\` | \`{exists: bool, mode, hint?}\` | \`params\` for placeholders. \`mode: 'filtered'|'any_rows'\` |
 | \`pg_get_indexes\` | \`{indexes, count, totalCount?}\` | Default \`limit: 100\` without \`table\`. Use \`schema\`/\`limit\` to filter. Index objects have \`name\`, \`type\`, \`columns\` |
 | \`pg_list_objects\` | \`{objects, count, totalCount, byType}\` | Use \`limit\` to cap results, \`type\`/\`types\` to filter |
-| \`pg_object_details\` | \`{name, schema, type, returnType?, ...}\` | Functions: \`returnType\` alias. Views/Mat. views: \`definition\` |
-| \`pg_analyze_db_health\` | \`{cacheHitRatio: {ratio, heap, index, status}}\` | \`ratio\` = primary numeric %. \`bloat\` available |
-| \`pg_describe_table\` | \`{columns, indexes, constraints, foreignKeys}\` | Columns include \`notNull\` (alias for \`!nullable\`), \`foreignKey\`. \`constraints\` includes PK, UNIQUE, CHECK, NOT NULL. ⚠️ \`rowCount: -1\` = no statistics (run ANALYZE) |
+| \`pg_object_details\` | \`{name, schema, type, returnType?, ...}\` | Functions: \`returnType\` alias. Views/Mat. views: \`definition\`. Tables: equivalent to \`pg_describe_table\` (columns, primaryKey, indexes, constraints, foreignKeys) |
+| \`pg_analyze_db_health\` | \`{cacheHitRatio, databaseSize, tableStats, unusedIndexes, tablesNeedingVacuum, connections, bloat, isReplica, overallScore, overallStatus}\` | \`cacheHitRatio\`: \`{ratio, heap, index, status}\`. \`overallStatus\`: \`healthy|needs_attention|critical\`. Optional sections via \`includeIndexes\`, \`includeVacuum\`, \`includeConnections\` |
+| \`pg_describe_table\` | \`{name, schema, type, owner, rowCount, columns, primaryKey, indexes, constraints, foreignKeys}\` | Columns include \`notNull\` (alias for \`!nullable\`), \`foreignKey\`. \`constraints\` includes PK, UNIQUE, CHECK, NOT NULL. ⚠️ \`rowCount: -1\` = stale/missing statistics (run ANALYZE on the table). Small tables (<~50 rows) may show -1 until first ANALYZE |
 | \`pg_analyze_query_indexes\` | \`{plan, issues, recommendations}\` | \`verbosity\`: 'summary' (default) or 'full'. Summary mode returns condensed plan |
-| \`pg_list_tables\` | \`{tables, count}\` | Use \`schema\` to filter, \`limit\` to cap results |
+| \`pg_list_tables\` | \`{tables, count, totalCount, truncated?, hint?}\` | Use \`schema\` to filter, \`limit\` to cap results, \`exclude\` to hide extension schemas (e.g., \`['cron', 'topology', 'partman']\`) |
 | List operations | \`{items, count}\` | Access via \`result.tables\`, \`result.views\`, etc. |
 | \`pg_jsonb_agg groupBy\` | \`{result: [{group_key, items}], count, grouped: true}\` | Without groupBy: \`{result: [...], count, grouped: false}\` |
 | \`pg_vector_aggregate\` | \`{average_vector, count}\` or \`{groups: [{group_key, average_vector, count}]}\` | Without/with \`groupBy\` |
+| \`pg_index_stats\` | \`{indexes, count, truncated?, totalCount?}\` | Default 50 rows. Use \`limit: 0\` for all |
+| \`pg_table_stats\` | \`{tables, count, truncated?, totalCount?}\` | Default 50 rows. Use \`limit: 0\` for all |
+| \`pg_vacuum_stats\` | \`{tables, count, truncated?, totalCount?}\` | Default 50 rows. Use \`limit: 0\` for all |
+| \`pg_stat_statements\` | \`{statements, count, truncated?, totalCount?}\` | Default 20 rows. \`orderBy\` supported |
+| \`pg_query_plan_stats\` | \`{queryPlanStats, count, truncated?, totalCount?}\` | Default 20 rows. \`truncateQuery: 0\` for full text |
+| \`pg_stat_activity\` | \`{connections, count}\` | \`includeIdle: true\` to include idle connections |
+| \`pg_locks\` | \`{locks}\` | \`showBlocked: true\` switches to blocked/blocking pid format |
+| \`pg_bloat_check\` | \`{tables, count}\` | Tables with \`live_tuples\`, \`dead_tuples\`, \`dead_pct\` |
+| \`pg_cache_hit_ratio\` | \`{heap_read, heap_hit, cache_hit_ratio}\` | All fields nullable (0 tables = null). Flat response, differs from \`pg_analyze_db_health.cacheHitRatio\` |
+| \`pg_seq_scan_tables\` | \`{tables, count, minScans, hint, truncated?, totalCount?}\` | Default 50 rows. \`minScans\` default: 10 |
+| \`pg_connection_pool_optimize\` | \`{current, config, waitEvents, recommendations}\` | No params needed |
+| \`pg_performance_baseline\` | \`{name, timestamp, metrics}\` | \`metrics\`: \`cache\`, \`tables\`, \`indexes\`, \`connections\`, \`databaseSize\` |
+| \`pg_duplicate_indexes\` | \`{duplicateIndexes, count, hint, truncated?, totalCount?}\` | Default 50 rows. \`duplicate_type\`: EXACT_DUPLICATE, OVERLAPPING, SUBSET |
+| \`pg_query_plan_compare\` | \`{query1, query2, analysis, fullPlans}\` | \`analysis.costDifference\` + \`recommendation\` |
+| \`pg_unused_indexes\` | \`{unusedIndexes, count, hint, truncated?, totalCount?}\` | Default 20 rows. \`summary: true\` → \`{summary, bySchema, totalCount}\` |
 
 ## API Mapping
 
@@ -80,10 +100,10 @@ export const SERVER_INSTRUCTIONS = `# postgres-mcp Code Mode
 - \`pg_vector_distance\`: Calculate distance between two vectors. \`metric\`: 'l2' (default), 'cosine', 'inner_product'. Returns \`{distance, metric}\`
 - \`pg_vector_cluster\`: \`clusters\` = \`k\`. Returns centroids with \`{preview, dimensions, truncated}\` format for large vectors (>10 dims)—use \`pg_vector_distance\` to assign rows
 - \`pg_vector_create_index\`: Use \`type\` (or alias \`method\`) with values 'ivfflat' or 'hnsw'. IVFFlat: \`lists\` param. HNSW: \`m\`, \`efConstruction\` params
-- \`pg_vector_performance\`: Auto-generates testVector from first row if omitted. Returns \`testVectorSource: 'auto-generated'|'user-provided'\`
+- \`pg_vector_performance\`: Auto-generates testVector from first row if omitted. Returns \`testVectorSource: 'auto-generated from first row'|'user-provided'\`
 - \`pg_vector_validate\`: Returns \`{valid: bool, vectorDimensions}\`. Empty vector \`[]\` returns \`{valid: true, vectorDimensions: 0}\`
 - ⛔ \`pg_vector_embed\`: Demo only (hash-based). Use OpenAI/Cohere for production.
-- \`pg_hybrid_search\`: Supports \`schema.table\` format (auto-parsed). Combines vector similarity and full-text search with weighted scoring. Code mode alias: \`pg.hybridSearch()\` → \`pg.vector.hybridSearch()\`
+- \`pg_hybrid_search\`: Supports \`schema.table\` format (auto-parsed). Combines vector similarity and full-text search with weighted scoring. \`textColumn\` auto-detects type: uses tsvector columns directly, wraps text columns with \`to_tsvector()\`. Code mode alias: \`pg.hybridSearch()\` → \`pg.vector.hybridSearch()\`
 - 📝 **Error Handling**: Vector tools return \`{success: false, error: "...", suggestion: "..."}\` for validation/semantic errors (dimension mismatch, non-vector column, table not found). Check \`success\` field before processing results.
 
 ## JSONB Tools
@@ -102,7 +122,6 @@ export const SERVER_INSTRUCTIONS = `# postgres-mcp Code Mode
 
 **Top-Level Aliases**: \`pg.jsonbExtract()\`, \`pg.jsonbSet()\`, \`pg.jsonbInsert()\`, \`pg.jsonbDelete()\`, \`pg.jsonbContains()\`, \`pg.jsonbPathQuery()\`, \`pg.jsonbAgg()\`, \`pg.jsonbObject()\`, \`pg.jsonbArray()\`, \`pg.jsonbKeys()\`, \`pg.jsonbStripNulls()\`, \`pg.jsonbTypeof()\`, \`pg.jsonbValidatePath()\`, \`pg.jsonbMerge()\`, \`pg.jsonbNormalize()\`, \`pg.jsonbDiff()\`, \`pg.jsonbIndexSuggest()\`, \`pg.jsonbSecurityScan()\`, \`pg.jsonbStats()\`
 
-
 ## Stats Tools
 
 - All stats tools support \`schema.table\` format (auto-parsed, embedded schema takes priority over explicit \`schema\` param)
@@ -117,11 +136,16 @@ export const SERVER_INSTRUCTIONS = `# postgres-mcp Code Mode
 
 **Top-Level Aliases**: \`pg.descriptive()\`, \`pg.percentiles()\`, \`pg.correlation()\`, \`pg.regression()\`, \`pg.timeSeries()\`, \`pg.distribution()\`, \`pg.hypothesis()\`, \`pg.sampling()\`
 
+## Text Tools
+
+- \`pg_text_search\`/\`pg_text_rank\`: Column must be \`text\` type—pre-built \`tsvector\` columns are **not** supported (wrap with \`to_tsvector()\` fails on tsvector input). Use \`pg_read_query\` with raw FTS SQL for tsvector columns
+- \`pg_create_fts_index\`: Returns \`{success, index, config, skipped}\`. \`skipped: true\` = index already existed (IF NOT EXISTS). \`ifNotExists\` defaults to \`true\`
+
 ## Performance Tools
 
 Core (20 methods): \`explain()\`, \`explainAnalyze()\`, \`explainBuffers()\`, \`indexStats()\`, \`tableStats()\`, \`statStatements()\`, \`statActivity()\`, \`locks()\`, \`bloatCheck()\`, \`cacheHitRatio()\`, \`seqScanTables()\`, \`indexRecommendations()\`, \`queryPlanCompare()\`, \`baseline()\`, \`connectionPoolOptimize()\`, \`partitionStrategySuggest()\`, \`unusedIndexes()\`, \`duplicateIndexes()\`, \`vacuumStats()\`, \`queryPlanStats()\`
 
-Wrappers (3): \`blockingQueries()\`→\`locks({showBlocked:true})\`, \`longRunningQueries({ seconds | minDuration }?)\` filters by duration (returns \`statActivity\` format), \`analyzeTable({ table })\` runs ANALYZE (accepts \`schema.table\` format)
+Wrappers (3): \`blockingQueries()\`→\`locks({showBlocked:true})\`, \`longRunningQueries({ seconds | minDuration }?)\` filters by duration (returns \`{longRunningQueries, count, threshold}\`), \`analyzeTable({ table })\` runs ANALYZE (accepts \`schema.table\` format)
 
 - \`explain({ sql, format?, params? })\`: Supports \`format: 'text'|'json'|'yaml'|'xml'\`. Default: text. Use \`params: [value]\` for \`$1, $2\` placeholders
 - \`explainAnalyze({ sql, format?, params? })\`: Same format/params options as explain
@@ -159,7 +183,7 @@ Core: \`databaseSize()\`, \`tableSizes()\`, \`connectionStats()\`, \`showSetting
 - \`recoveryStatus()\`: Returns \`{in_recovery: boolean, last_replay_timestamp: string|null}\`
 - \`replicationStatus()\`: Returns \`{role: 'primary'|'replica', replicas: [...]}\` for primary, or \`{role: 'replica', replay_lag, ...}\` for replica
 - \`resourceUsageAnalyze()\`: Returns \`{backgroundWriter, checkpoints, connectionDistribution, bufferUsage, activity, analysis}\` with all counts as numbers
-- \`alertThresholdSet({metric?: 'connection_usage'})\`: Returns recommended thresholds. ⛔ Invalid metric throws validation error. Valid metrics: connection_usage, cache_hit_ratio, replication_lag, dead_tuples, long_running_queries, lock_wait_time
+- \`alertThresholdSet({metric?: 'connection_usage'})\`: Returns recommended thresholds. Invalid metric returns \`{success: false, error: "..."}\`. Valid metrics: connection_usage, cache_hit_ratio, replication_lag, dead_tuples, long_running_queries, lock_wait_time
 
 📦 **AI-Optimized Payloads**: Tools return limited results by default to reduce context size:
 - \`tableSizes({ limit? })\`: Default 50 rows. Returns \`truncated: true\` + \`totalCount\` when limited. Use \`limit: 0\` for all
@@ -185,6 +209,17 @@ Aliases: \`tableName\`→\`table\`, \`indexName\`→\`index\`, \`param\`/\`setti
 
 **Top-Level Aliases**: \`pg.vacuum()\`, \`pg.vacuumAnalyze()\`, \`pg.analyze()\`, \`pg.reindex()\`, \`pg.cluster()\`, \`pg.setConfig()\`, \`pg.reloadConf()\`, \`pg.resetStats()\`, \`pg.cancelBackend()\`, \`pg.terminateBackend()\`
 
+**Discovery**: \`pg.admin.help()\` returns \`{methods, methodAliases, examples}\` object
+
+**Response structures**:
+- \`vacuum()\` / \`vacuumAnalyze()\`: \`{success, message, table?, schema?, hint?}\` (hint present when verbose: true)
+- \`analyze()\`: \`{success, message, table?, schema?, columns?}\`
+- \`reindex()\`: \`{success, message}\`
+- \`cluster()\`: \`{success, message, table?, index?}\` (table/index present for table-specific cluster)
+- \`setConfig()\`: \`{success, message, parameter, value}\`
+- \`reloadConf()\` / \`resetStats()\`: \`{success, message}\`
+- \`cancelBackend()\` / \`terminateBackend()\`: \`{success, message}\`
+
 ## Backup Tools
 
 Core: \`dumpTable()\`, \`dumpSchema()\`, \`copyExport()\`, \`copyImport()\`, \`createBackupPlan()\`, \`restoreCommand()\`, \`physical()\`, \`restoreValidate()\`, \`scheduleOptimize()\`
@@ -195,7 +230,7 @@ Response Structures:
 - \`copyImport\`: \`{command, stdinCommand, notes}\` — Both file and stdin COPY commands
 - \`createBackupPlan\`: \`{strategy: {fullBackup, walArchiving}, estimates}\`
 - \`restoreCommand\`: \`{command, warnings?, notes}\` — Warnings when \`database\` omitted
-- \`restoreValidate\`: \`{validationSteps: [{step, name, command?, commands?, note?}], recommendations}\` — Note: \`note\` field only for pg_dump default type
+- \`restoreValidate\`: \`{note?, validationSteps: [{step, name, command?, commands?, note?}], recommendations}\` — Top-level \`note\` when \`backupType\` omitted (defaults to pg_dump). Step-level \`note\` for non-command steps
 - \`physical\`: \`{command, notes, requirements}\`
 - \`scheduleOptimize\`: \`{analysis, recommendation, commands}\`
 
@@ -207,7 +242,7 @@ Response Structures:
 - \`pg_copy_import\`: Generates COPY FROM command. Supports \`schema.table\` format (auto-parsed, takes priority over \`schema\` param). \`columns\` array, \`filePath\`, \`format\`, \`header\`, \`delimiter\`
 - \`pg_restore_command\`: Include \`database\` parameter for complete command. Optional \`schemaOnly\`, \`dataOnly\`
 - \`pg_create_backup_plan\`: Generates backup strategy with cron schedule. \`frequency\`: 'hourly'|'daily'|'weekly', \`retention\` count
-- \`pg_backup_physical\`: Generates pg_basebackup command. \`format\`: 'plain'|'tar', \`checkpoint\`: 'fast'|'spread', \`compress\`: 0-9
+- \`pg_backup_physical\`: Generates pg_basebackup command. \`format\`: 'plain'|'tar' (default: 'tar'), \`checkpoint\`: 'fast'|'spread', \`compress\`: 0-9
 - \`pg_restore_validate\`: Generates validation commands. \`backupType\`: 'pg_dump' (default)|'pg_basebackup'
 - \`pg_backup_schedule_optimize\`: Analyzes database activity patterns and recommends optimal backup schedule
 
@@ -219,14 +254,13 @@ Defaults: \`threshold\`=0.3 (use 0.1-0.2 for partial), \`maxDistance\`=3 (use 5+
 
 - All text tools support \`schema.table\` format (auto-parsed, embedded schema takes priority over explicit \`schema\` param)
 - \`pg_text_search\`: Supports both \`column\` (singular string) and \`columns\` (array). Either is valid—\`column\` auto-converts to array
-- \`pg_trigram_similarity\` vs \`pg_similarity_search\`: Both use pg_trgm. First filters by threshold; second uses set_limit() with %
+- \`pg_trigram_similarity\`, \`pg_fuzzy_match\`, \`pg_regexp_match\`, \`pg_like_search\`: All default to 100 results to prevent large payloads. Use \`limit: 0\` for all rows
 - \`pg_fuzzy_match\`: Levenshtein returns distance (lower=better). Soundex/metaphone return phonetic codes (exact match only). ⛔ Invalid \`method\` values throw error with valid options
 - \`pg_text_normalize\`: Removes accents only (unaccent). Does NOT lowercase/trim
 - 📍 **Table vs Standalone**: \`normalize\`, \`sentiment\`, \`toVector\`, \`toQuery\`, \`searchConfig\` are standalone (text input only). For phonetic matching: use \`pg_fuzzy_match\` with \`method: 'soundex'|'metaphone'\` (direct MCP), or \`pg.text.soundex()\`/\`pg.text.metaphone()\` (Code Mode convenience wrappers that call fuzzyMatch internally)
 
 **Top-Level Aliases**: \`pg.textSearch()\`, \`pg.textRank()\`, \`pg.textHeadline()\`, \`pg.textNormalize()\`, \`pg.textSentiment()\`, \`pg.textToVector()\`, \`pg.textToQuery()\`, \`pg.textSearchConfig()\`, \`pg.textTrigramSimilarity()\`, \`pg.textFuzzyMatch()\`, \`pg.textLikeSearch()\`, \`pg.textRegexpMatch()\`, \`pg.textCreateFtsIndex()\`
 
-
 ## Schema Tools
 
 Core: \`listSchemas()\`, \`createSchema()\`, \`dropSchema()\`, \`listViews()\`, \`createView()\`, \`dropView()\`, \`listSequences()\`, \`createSequence()\`, \`dropSequence()\`, \`listFunctions()\`, \`listTriggers()\`, \`listConstraints()\`
@@ -235,18 +269,17 @@ Response Structures:
 - \`listSchemas()\`: \`{schemas: string[], count}\`
 - \`listViews({ includeMaterialized?, truncateDefinition?, limit? })\`: \`{views: [{schema, name, type, definition, definitionTruncated?}], count, hasMatViews, truncatedDefinitions?, truncated, note?}\`. Default \`limit: 50\` (use \`0\` for all). Default \`truncateDefinition: 500\` chars (use \`0\` for full definitions). \`truncated\` always included (\`true\`/\`false\`)
 - \`listSequences({ schema? })\`: \`{sequences: [{schema, name, owned_by}], count}\`. Note: \`owned_by\` omits \`public.\` prefix for sequences in public schema (e.g., \`users.id\` not \`public.users.id\`)
-- \`listFunctions({ schema?, limit?, exclude? })\`: \`{functions: [{schema, name, arguments, returns, language, volatility}], count, limit, note?}\`
+- \`listFunctions({ schema?, limit?, exclude? })\`: \`{functions: [{schema, name, arguments, returns, language, volatility}], count, limit}\`
 - \`listTriggers({ schema?, table? })\`: \`{triggers: [{schema, table_name, name, timing, events, function_name, enabled}], count}\`
 - \`listConstraints({ schema?, table?, type? })\`: \`{constraints: [{schema, table_name, name, type, definition}], count}\`. Type codes: \`p\`=primary_key, \`f\`=foreign_key, \`u\`=unique, \`c\`=check
 - \`dropSchema/dropView/dropSequence\`: All return \`{existed: true/false}\` to indicate if object existed before drop
-- \`createSchema/createSequence\` (with \`ifNotExists\`) and \`createView\` (with \`orReplace\`): Return \`{alreadyExisted: true/false}\` to indicate if object existed before creation
+- \`createSchema/createSequence\` (with \`ifNotExists\`) and \`createView\` (with \`orReplace\`): Return \`{alreadyExisted: true/false}\` when the flag is set. Without \`ifNotExists\`/\`orReplace\`, the field is omitted
 
 - \`pg_create_view\`: Supports \`schema.name\` format (auto-parsed). Use \`orReplace: true\` for CREATE OR REPLACE. \`checkOption\`: 'cascaded', 'local', 'none'. ⛔ OR REPLACE can add new columns but cannot rename/remove existing ones—PostgreSQL limitation
 - \`pg_create_sequence\`: Supports \`schema.name\` format. Parameters: \`start\`, \`increment\`, \`minValue\`, \`maxValue\`, \`cache\`, \`cycle\`, \`ownedBy\`, \`ifNotExists\`
-- \`pg_list_functions\`: Default limit=500. Use \`schema: 'public'\`, \`limit: 2000\`, or \`exclude: ['postgis']\` to filter. ⚠️ \`exclude\` filters by **schema name** AND extension-owned functions. Note: Aggressive \`exclude\` may return 0 results if all functions belong to excluded extensions
-
-**Discovery**: \`pg.schema.help()\` returns \`{methods: string[], examples: string[]}\` object with available methods and usage examples
+- \`pg_list_functions\`: Default limit=500. Use \`schema: 'public'\`, \`limit: 2000\`, or \`exclude: ['postgis', 'pg_trgm', 'ltree', 'citext', 'fuzzystrmatch', 'pg_stat_statements', 'hypopg', 'unaccent', 'pg_stat_kcache', 'pgcrypto', 'partman', 'vector', 'topology']\` to filter. ⚠️ \`exclude\` filters by **schema name** AND extension-owned functions. The \`language\` filter does NOT exclude extension functions—use \`exclude\` alongside \`language\` for clean results. Note: Aggressive \`exclude\` may return 0 results if all functions belong to excluded extensions
 
+**Discovery**: \`pg.schema.help()\` returns \`{methods, methodAliases, examples}\` object
 
 ## Partitioning Tools
 
@@ -256,6 +289,8 @@ Response Structures:
 - \`pg_list_partitions\`: Default \`limit: 50\` (use \`0\` for all). Returns \`{partitions, count, truncated, totalCount?}\`. Uses \`bounds\` field (consistent with \`pg_partition_info\`)
 - \`pg_partition_info\`: Returns \`{tableInfo, partitions, totalSizeBytes}\`. Uses \`bounds\` field
 - Both list/info tools support \`schema.table\` format (auto-parsed) and accept \`table\`, \`parent\`, \`parentTable\`, or \`name\` aliases
+- Response structures: \`pg_create_partitioned_table\` → \`{success, table, partitionBy, partitionKey, primaryKey?}\`. \`pg_create_partition\` → \`{success, partition, parent, bounds, subpartitionBy?, subpartitionKey?}\`. \`pg_attach_partition\` → \`{success, parent, partition, bounds}\`. \`pg_detach_partition\` → \`{success, parent, partition}\`
+- ⚠️ Sub-partitioning: \`subpartitionBy\`/\`subpartitionKey\` on \`pg_create_partition\` makes a partition itself partitionable. The parent's \`primaryKey\` must include the sub-partition key column (PostgreSQL constraint)
 - 📍 Code Mode: \`pg.partitioning.create()\` = \`createPartition\`, NOT \`createPartitionedTable\`
 
 ## pg_partman Tools
@@ -266,7 +301,7 @@ Response Structures:
 - \`pg_partman_show_partitions\`: Default \`limit: 50\` (use \`0\` for all). Returns \`truncated\` + \`totalCount\` when limited. \`parentTable\` required. Supports \`schema.table\` format (auto-parsed)
 - \`pg_partman_check_default\`/\`partition_data\`: \`parentTable\` required. Supports \`schema.table\` format (auto-parsed)
 - \`pg_partman_set_retention\`: ⚠️ **CAUTION: Default is DROP** — \`retentionKeepTable: false\` (default) = DROP partitions, \`true\` = detach only (safer). Pass \`retention: null\` to disable retention
-- \`pg_partman_undo_partition\`: \`targetTable\` MUST exist before calling. Requires both \`parentTable\` and \`targetTable\`/\`target\`
+- \`pg_partman_undo_partition\`: \`targetTable\` MUST exist before calling. Requires both \`parentTable\` and \`targetTable\`/\`target\`. ⚠️ Parent table and child partitions remain after undo—use \`DROP TABLE parent CASCADE\` to clean up
 - \`pg_partman_analyze_partition_health\`: Default \`limit: 50\` (use \`0\` for all). Returns \`truncated\` + \`totalCount\` when limited. \`summary.overallHealth\`: 'healthy'|'warnings'|'issues_found'
 - 📝 **Schema Resolution**: All partman tools auto-prefix \`public.\` when no schema specified in \`parentTable\`
 - 📝 **Aliases**: \`parentTable\` accepts \`table\`, \`parent\`, \`name\`. \`controlColumn\` accepts \`control\`, \`column\`. \`targetTable\` accepts \`target\`
@@ -275,11 +310,11 @@ Response Structures:
 
 Core: \`createExtension()\`, \`queryStats()\`, \`topCpu()\`, \`topIo()\`, \`databaseStats()\`, \`resourceAnalysis()\`, \`reset()\`
 
-- \`pg_kcache_query_stats\`: Default \`limit: 50\` (use \`0\` for all). Returns \`truncated\` + \`totalCount\` when limited. \`orderBy\`: 'total_time' (default), 'cpu_time', 'reads', 'writes'. \`queryPreviewLength\`: chars for query preview (default: 100, max: 500, 0 for full). ⛔ 'calls' NOT valid for orderBy—use \`minCalls\` param
-- \`pg_kcache_resource_analysis\`: Default \`limit: 50\` (use \`0\` for all). Returns \`truncated\` + \`totalCount\` when limited. \`minCalls\`, \`queryPreviewLength\` supported. Classifies queries as 'CPU-bound', 'I/O-bound', or 'Balanced'
-- \`pg_kcache_top_cpu\`: Top CPU-consuming queries. \`limit\` param (default: 10)
-- \`pg_kcache_top_io\`: \`type\`/\`ioType\` (alias): 'reads', 'writes', 'both' (default). \`limit\` param (default: 10)
-- \`pg_kcache_database_stats\`: Aggregated CPU/IO stats per database
+- \`pg_kcache_query_stats\`: Default \`limit: 20\` (use \`0\` for all). Returns \`truncated\` + \`totalCount\` when limited. \`orderBy\`: 'total_time' (default), 'cpu_time', 'reads', 'writes'. \`queryPreviewLength\`: chars for query preview (default: 100, max: 500, 0 for full). ⛔ 'calls' NOT valid for orderBy—use \`minCalls\` param
+- \`pg_kcache_resource_analysis\`: Default \`limit: 20\` (use \`0\` for all). Returns \`truncated\` + \`totalCount\` when limited. \`minCalls\`, \`queryPreviewLength\` supported. Classifies queries as 'CPU-bound', 'I/O-bound', or 'Balanced'
+- \`pg_kcache_top_cpu\`: Top CPU-consuming queries. \`limit\` param (default: 10). \`queryPreviewLength\`: chars for query preview (default: 100, max: 500, 0 for full). Returns \`truncated\` + \`totalCount\` when limited
+- \`pg_kcache_top_io\`: \`type\`/\`ioType\` (alias): 'reads', 'writes', 'both' (default). \`limit\` param (default: 10). \`queryPreviewLength\`: chars for query preview (default: 100, max: 500, 0 for full). Returns \`truncated\` + \`totalCount\` when limited
+- \`pg_kcache_database_stats\`: Aggregated CPU/IO stats per database. Optional \`database\` param to filter specific db
 - \`pg_kcache_reset\`: Resets pg_stat_kcache AND pg_stat_statements statistics
 
 ## citext Tools
@@ -308,7 +343,7 @@ Core: \`createExtension()\`, \`query()\`, \`match()\`, \`subpath()\`, \`lca()\`,
 - \`pg_ltree_convert_column\`: Convert TEXT column to ltree. Supports \`schema.table\` format. \`col\` alias for \`column\`. Returns \`{previousType}\`. ⚠️ When views depend on column, returns \`{success: false, dependentViews, hint}\`—drop/recreate views manually
 - \`pg_ltree_create_index\`: Create GiST index on ltree column. Supports \`schema.table\` format. Auto-generates index name if \`indexName\` omitted. Returns \`{indexName, indexType: 'gist', alreadyExists?}\`
 
-**Discovery**: \`pg.ltree.help()\` returns \`{methods, aliases, examples}\` object. Top-level aliases available: \`pg.ltreeQuery()\`, \`pg.ltreeMatch()\`, etc.
+**Discovery**: \`pg.ltree.help()\` returns \`{methods, methodAliases, examples}\` object. Top-level aliases available: \`pg.ltreeQuery()\`, \`pg.ltreeMatch()\`, etc.
 
 ## PostGIS Tools
 
@@ -325,7 +360,7 @@ Core: \`createExtension()\`, \`query()\`, \`match()\`, \`subpath()\`, \`lca()\`,
 
 **Geometry Operations (Table-based):**
 - \`pg_buffer\`: Create buffer zone around table geometries. Default limit: 50 rows. Default simplify: 10m (set \`simplify: 0\` to disable). Returns \`truncated: true\` + \`totalCount\` when results are truncated. Use \`limit: 0\` for all rows
-- \`pg_geo_transform\`: Transform table geometries between SRIDs. Default limit: 50 rows. Returns \`truncated: true\` + \`totalCount\` when results are truncated. Use \`limit: 0\` for all rows. \`fromSrid\`/\`sourceSrid\` and \`toSrid\`/\`targetSrid\` aliases
+- \`pg_geo_transform\`: Transform table geometries between SRIDs. Default limit: 50 rows. Returns \`truncated: true\` + \`totalCount\` when results are truncated. Use \`limit: 0\` for all rows. Auto-detects \`fromSrid\` from column metadata if not provided (returns \`autoDetectedSrid: true\`). \`fromSrid\`/\`sourceSrid\` and \`toSrid\`/\`targetSrid\` aliases
 - \`pg_geo_cluster\`: Spatial clustering (DBSCAN/K-Means). K-Means: If \`numClusters\` exceeds row count, automatically clamps to available rows with \`warning\` field. DBSCAN: Returns contextual \`hints\` array explaining parameter effects (e.g., "All points formed single cluster—decrease eps") and \`parameterGuide\` explaining eps/minPoints trade-offs
 
 **Geometry Operations (Standalone WKT/GeoJSON):**
@@ -337,7 +372,7 @@ Core: \`createExtension()\`, \`query()\`, \`match()\`, \`subpath()\`, \`lca()\`,
 - \`pg_postgis_create_extension\`: Enable PostGIS extension (idempotent)
 - \`pg_geo_index_optimize\`: Analyze spatial indexes. Without \`table\` param, analyzes all spatial indexes
 
-**Code Mode Aliases:** \`pg.postgis.addColumn()\` → \`geometryColumn\`, \`pg.postgis.indexOptimize()\` → \`geoIndexOptimize\`. Note: \`pg.{group}.help()\` returns \`{methods, aliases, examples}\`
+**Code Mode Aliases:** \`pg.postgis.addColumn()\` → \`geometryColumn\`, \`pg.postgis.indexOptimize()\` → \`geoIndexOptimize\`, \`pg.postgis.geoCluster()\` → \`pg_geo_cluster\`, \`pg.postgis.geoTransform()\` → \`pg_geo_transform\`. Note: \`pg.{group}.help()\` returns \`{methods, methodAliases, examples}\`
 
 ## Cron Tools (pg_cron)
 
@@ -346,13 +381,13 @@ Core: \`createExtension()\`, \`schedule()\`, \`scheduleInDatabase()\`, \`unsched
 - \`pg_cron_schedule\`: Schedule a cron job. \`schedule\` supports standard cron (\`0 5 * * *\`) or interval (\`1 second\` to \`59 seconds\`). ⚠️ Interval syntax only works for 1-59 seconds—for 60+ seconds, use cron syntax (e.g., \`* * * * *\` for every minute). Use \`name\`/\`jobName\` for identification. \`command\`/\`sql\`/\`query\` aliases supported. Note: pg_cron allows duplicate job names; use unique names to avoid confusion when unscheduling
 - \`pg_cron_schedule_in_database\`: Schedule job in specific database. \`database\`/\`db\` aliases. Optional \`username\`, \`active\` params
 - \`pg_cron_unschedule\`: Remove job by \`jobId\` or \`jobName\`. If both provided, \`jobName\` takes precedence (with warning)
-- \`pg_cron_alter_job\`: Modify existing job. Can change \`schedule\`, \`command\`, \`database\`, \`username\`, \`active\`. ⛔ Non-existent jobId throws error
+- \`pg_cron_alter_job\`: Modify existing job. Can change \`schedule\`, \`command\`, \`database\`, \`username\`, \`active\`. ⛔ Non-existent jobId returns error
 - \`pg_cron_list_jobs\`: List all jobs. Default \`limit: 50\` (use \`0\` for all). Optional \`active\` boolean filter. Returns \`truncated\` + \`totalCount\` when limited. Returns \`hint\` when jobs have no name
 - \`pg_cron_job_run_details\`: View execution history. Default \`limit: 50\`. Optional \`jobId\`, \`status\` ('running'|'succeeded'|'failed') filters. Returns \`truncated\` + \`totalCount\` when limited. Returns \`summary\` with counts
 - \`pg_cron_cleanup_history\`: Delete old run records. \`olderThanDays\`/\`days\` param (default: 7). Optional \`jobId\` to target specific job
 - \`pg_cron_create_extension\`: Enable pg_cron extension (idempotent). Requires superuser
 
-**Discovery**: \`pg.cron.help()\` returns \`{methods, aliases, examples}\` object
+**Discovery**: \`pg.cron.help()\` returns \`{methods, methodAliases, examples}\` object
 
 ## pgcrypto Tools
 
@@ -372,7 +407,7 @@ Core: \`createExtension()\`, \`hash()\`, \`hmac()\`, \`encrypt()\`, \`decrypt()\
 
 **Top-Level Aliases**: \`pg.pgcryptoHash()\`, \`pg.pgcryptoEncrypt()\`, \`pg.pgcryptoDecrypt()\`, \`pg.pgcryptoGenRandomUuid()\`, etc.
 
-**Discovery**: \`pg.pgcrypto.help()\` returns \`{methods, aliases, examples}\` object
+**Discovery**: \`pg.pgcrypto.help()\` returns \`{methods, methodAliases, examples}\` object
 
 ## Code Mode Sandbox
 
@@ -400,6 +435,13 @@ Core: \`begin()\`, \`commit()\`, \`rollback()\`, \`savepoint()\`, \`rollbackTo()
   - **Join existing**: With \`transactionId\`/\`tx\`/\`txId\`—no auto-commit, caller controls via commit/rollback
 - \`statements\`: Array of \`{sql: "...", params?: [...]}\` objects. ⚠️ Each object MUST have \`sql\` key
 - \`isolationLevel\`: Optional isolation level for new transactions ('READ COMMITTED', 'REPEATABLE READ', 'SERIALIZABLE')
+- Supports SELECT statements inside \`statements\`—results include \`rows\` in the response for mixed read/write workflows
+
+**Aborted Transaction State:**
+- ⚠️ If any statement in a transaction fails, PostgreSQL puts the transaction into an **aborted state**
+- In aborted state, only \`ROLLBACK\` or \`ROLLBACK TO SAVEPOINT\` commands are accepted—all other commands will error
+- Use \`pg_transaction_rollback\` to end the transaction, or \`pg_transaction_rollback_to\` to recover to a savepoint
+- \`pg_transaction_commit\` on an aborted transaction will detect the state and report it (not silently rollback)
 
 **Response Structures:**
 - \`begin\`: \`{transactionId, isolationLevel: 'READ COMMITTED', message}\`

package/dist/constants/ServerInstructions.js.map

@@ -1 +1 @@
-{"version":3,"file":"ServerInstructions.js","sourceRoot":"","sources":["../../src/constants/ServerInstructions.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AACH,MAAM,CAAC,MAAM,mBAAmB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;yFA8YsD,CAAC"}
+{"version":3,"file":"ServerInstructions.js","sourceRoot":"","sources":["../../src/constants/ServerInstructions.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AACH,MAAM,CAAC,MAAM,mBAAmB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;yFAwbsD,CAAC"}