@kadoa/mcp 0.5.20 → 0.5.21

package/README.md

@@ -127,9 +127,10 @@ to retrieve the extracted records and display them as a table.
 > You: Use my "Product Scraper" template to scrape https://example-shop.com.
 
 Claude calls list_templates to find the matching template, then
-create_workflow with `templateId` and `urls` only — the prompt and
-schema are inherited from the template version. Returns the workflow
-ID for follow-up with get_workflow or fetch_data.
+create_workflow with `templateId` and `urls` only - never copy the prompt,
+entity, or schema into standalone creation. Those values are inherited from
+the template version. Returns the workflow ID for follow-up with get_workflow
+or fetch_data.
 ```
 
 ### Update a workflow and re-run
@@ -148,8 +149,10 @@ changes, and shows the updated field list.
 
 > You: Run it again with the new schema.
 
-Claude calls run_workflow and waits for completion, then fetches
-the latest data with fetch_data so you can verify the changes.
+Claude calls run_workflow and returns the workflow ID while the run proceeds
+asynchronously. Check back later with get_workflow or fetch_data. If a
+one-time or scheduled workflow fails, call run_workflow again with the same
+workflow ID and configuration - never delete and recreate it just to retry.
 ```
 
 ### Update deterministic workflow settings
@@ -247,14 +250,16 @@ Relevant files in `kadoa-backend`:
 
 1. Merge PRs to `main` using Conventional Commits (`feat:`, `fix:`, etc.). Release Please opens/maintains a `chore(main): release mcp x.y.z` PR.
 2. Merge the release PR. The [`release-please.yml`](.github/workflows/release-please.yml) workflow tags, drafts a GitHub Release, and publishes to npm (`latest` dist-tag).
-3. In `kadoa-backend`, bump `infra/docker/mcp/package.json` `@kadoa/mcp` to the new version, run `bun install` to refresh `bun.lock`, open a PR.
-4. Merge to `main`. CI (`main-build-deploy.yml`) builds and pushes `europe-west3-docker.pkg.dev/oceanic-base-310208/kadoa-artifacts/mcp-server:<IMAGE_TAG>` (tag shown in the build summary).
+3. After npm publication, `release-please.yml` opens or updates a deterministic PR in `kadoa-backend` that bumps `infra/docker/mcp/package.json` and refreshes `bun.lock`. The PR validates a frozen production install and Docker image build, and remains manually mergeable.
+4. Merge the backend PR to `main`. CI (`main-build-deploy.yml`) builds and pushes `europe-west3-docker.pkg.dev/oceanic-base-310208/kadoa-artifacts/mcp-server:<IMAGE_TAG>` (tag shown in the build summary).
 5. Trigger the **Deploy to Production** workflow ([`deploy-prod.yml`](https://github.com/kadoa-org/kadoa-backend/actions/workflows/deploy-prod.yml)) with:
    - **Target cluster:** `gcp`
    - **Deployment scope:** `mcp`
    - **Image tag:** the tag from step 4
    - **Method:** `kubectl`
 
+The cross-repository bump requires the `BACKEND_REPO_TOKEN` secret in this repository. The selected authentication approach is a fine-grained PAT scoped only to `kadoa-org/kadoa-backend` with Contents read/write and Pull requests read/write permissions (Metadata read is automatic). If the secret is not provisioned, the release workflow skips the backend bump with a warning rather than failing npm releases; an administrator must provision it before relying on automatic hosted-server updates.
+
 ### RC / test release
 
 Use this when you want to validate a change end-to-end against real clients (Claude Desktop, Cursor, ChatGPT) before promoting to `latest` / prod. The flow mirrors the prod one, but every step targets `rc` channels.

package/dist/index.js

@@ -52699,7 +52699,7 @@ function extractApiMessage(responseBody) {
   if (body.validationErrors && typeof body.validationErrors === "object" && body.validationErrors !== null) {
     const details = Object.entries(body.validationErrors).map(([field, err]) => `${field}: "${err}"`).join(", ");
     if (details) {
-      msg = msg ? `${msg} — Details: ${details}` : `Validation failed — Details: ${details}`;
+      msg = msg ? `${msg} - Details: ${details}` : `Validation failed - Details: ${details}`;
     }
   }
   if (!msg && Array.isArray(body.issues)) {
@@ -52709,7 +52709,7 @@ function extractApiMessage(responseBody) {
       return path ? `${path}: "${message}"` : `"${message}"`;
     }).join(", ");
     if (details)
-      msg = `Validation failed — Details: ${details}`;
+      msg = `Validation failed - Details: ${details}`;
   }
   return msg;
 }
@@ -52895,14 +52895,14 @@ function registerTools(server, ctx, capabilities) {
     });
   }));
   const urlInputShape = {
-    url: exports_external.string().optional().describe("Single URL — prefer using 'urls' instead. If both are provided, 'urls' takes precedence."),
+    url: exports_external.string().optional().describe("Single URL - prefer using 'urls' instead. If both are provided, 'urls' takes precedence."),
     urls: exports_external.preprocess(coerceArray(true), exports_external.array(exports_external.string()).min(1)).optional().describe("Starting URLs for the workflow (array of strings). Also accepts a single URL string.")
   };
   const extractionInputShape = {
     prompt: exports_external.string().optional().describe('Natural language description of what to extract (e.g., "Extract product prices and names"). Required unless templateId is provided.'),
     name: exports_external.string().optional().describe("Optional name for the workflow"),
     entity: exports_external.string().optional().describe("Entity name for extraction (e.g., 'Product', 'Job Posting')"),
-    schema: exports_external.preprocess(coerceArray(), exports_external.array(exports_external.object(SchemaFieldShape).strict())).optional().describe("Extraction schema fields. If omitted, the AI agent auto-detects the schema. When you do supply fields, mark the one that identifies a record with isKey — change detection needs it.")
+    schema: exports_external.preprocess(coerceArray(), exports_external.array(exports_external.object(SchemaFieldShape).strict())).optional().describe("Extraction schema fields. If omitted, the AI agent auto-detects the schema. When you do supply fields, mark the one that identifies a record with isKey - change detection needs it.")
   };
   const webhookAuthShape = exports_external.object({
     type: exports_external.enum(["bearer", "basic", "header"]).describe("Authentication type"),
@@ -53030,15 +53030,15 @@ function registerTools(server, ctx, capabilities) {
     return channels;
   }
   server.registerTool("create_workflow", {
-    description: "IMPORTANT: One workflow = one source. A single workflow can extract many different fields, tables, and sections from the same URL(s) using a rich schema. " + `Do NOT create separate workflows for different data points on the same page — instead, define one workflow with a multi-field schema covering everything needed.
+    description: "IMPORTANT: One workflow = one source. A single workflow can extract many different fields, tables, and sections from the same URL(s) using a rich schema. " + `Do NOT create separate workflows for different data points on the same page - instead, define one workflow with a multi-field schema covering everything needed.
 
 ` + "Create a data extraction workflow using agentic navigation. Supports one-time or scheduled runs. " + "If entity and schema are provided, they guide the extraction; otherwise the AI agent auto-detects the schema from the page. " + "The workflow runs asynchronously and may take several minutes. Do NOT poll or sleep-wait for completion. " + `Return the workflow ID to the user and let them check back later with get_workflow or fetch_data.
 
-` + "PREFER TEMPLATES: If the user's request matches an existing template, instantiate it via `templateId` instead of writing a fresh prompt/schema. " + "Use `list_templates` to discover available templates and `get_template` to inspect schemas before deciding. " + "When `templateId` is set, only `urls` is required — `prompt`, `entity`, and `schema` must NOT be supplied; they are inherited from the template version.\n\n" + "NOTE: This tool is for one-time or scheduled extraction ONLY. " + "For continuous Julie realtime monitoring (watching a page for data changes and alerting), use the create_realtime_monitor tool instead.",
+` + "PREFER TEMPLATES: If the user's request matches an existing template, instantiate it via `templateId` instead of writing a fresh prompt/schema. " + "Use `list_templates` to discover available templates and `get_template` to inspect schemas before deciding. " + "When creating from a template, call this tool with the template's `templateId`, the source `urls`, and optional `templateVersion` only. Never copy the template prompt, entity, or schema into standalone creation, and never silently match or rewrite inline configuration.\n\n" + "NOTE: This tool is for one-time or scheduled extraction ONLY. " + "For continuous Julie realtime monitoring (watching a page for data changes and alerting), use the create_realtime_monitor tool instead.",
     inputSchema: strictSchema({
       ...extractionInputShape,
       ...urlInputShape,
-      templateId: exports_external.string().optional().describe("Instantiate this workflow from a published template. When set, only 'urls' is required — prompt/entity/schema must NOT be supplied; they are inherited from the template version, and the workflow's output conforms to the template's declared schema (field names are enforced, not drifted). Discover templates via list_templates."),
+      templateId: exports_external.string().optional().describe("Instantiate this workflow from a published template. Pass the templateId, source urls, and optional templateVersion. Do not copy prompt/entity/schema from the template into this call - they are inherited from the published template version, and the workflow's output conforms to its declared schema. Discover templates via list_templates and get_template."),
       templateVersion: exports_external.preprocess(coerceNumber(), exports_external.number()).optional().describe("Specific published template version (integer) to instantiate. Defaults to the latest published version when templateId is set."),
       description: exports_external.string().max(500).optional().describe("Description of what this workflow does (max 500 characters)"),
       tags: exports_external.preprocess(coerceArray(true), exports_external.array(exports_external.string())).optional().describe("Tags for organizing workflows"),
@@ -53079,7 +53079,7 @@ function registerTools(server, ctx, capabilities) {
     let workflowId;
     if (args.templateId) {
       if (args.prompt || args.entity || args.schema) {
-        return errorResult("When 'templateId' is set, 'prompt', 'entity', and 'schema' must NOT be supplied — they are inherited from the template version.");
+        return errorResult("When 'templateId' is set, 'prompt', 'entity', and 'schema' must NOT be supplied - they are inherited from the template version.");
       }
       const { id } = await ctx.client.workflow.create({
         urls,
@@ -53235,7 +53235,7 @@ function registerTools(server, ctx, capabilities) {
     });
   }));
   server.registerTool("get_workflow", {
-    description: "Get canonical workflow details, including Assistant/session linkage, extraction intent, template-controlled parts, stale-data state, job/run state, schedule timezone, location, monitoring, validation, and realtime health. For Assistant-built workflows, extractionSpec—not the legacy prompt—is authoritative.",
+    description: "Get canonical workflow details, including Assistant/session linkage, extraction intent, template-controlled parts, stale-data state, job/run state, schedule timezone, location, monitoring, validation, and realtime health. For Assistant-built workflows, extractionSpec - not the legacy prompt - is authoritative.",
     inputSchema: {
       workflowId: exports_external.string().describe("The workflow ID")
     },
@@ -53599,7 +53599,7 @@ function registerTools(server, ctx, capabilities) {
     });
   }));
   server.registerTool("list_workflow_runs", {
-    description: "List a workflow's execution run history — each run's status, start/finish time, " + "record count, and errors. Use to answer 'when did this last succeed?' " + "(status='success', limit=1) or 'recent success/failure pattern?'. Distinct from " + "get_workflow_history, which is the config audit log. `status` is normally " + "success | failed | in_progress; a run whose backend state this server does not " + "recognize reports that raw state verbatim, so treat any other value as unknown " + "rather than as a failure.",
+    description: "List a workflow's execution run history - each run's status, start/finish time, " + "record count, and errors. Use to answer 'when did this last succeed?' " + "(status='success', limit=1) or 'recent success/failure pattern?'. Distinct from " + "get_workflow_history, which is the config audit log. `status` is normally " + "success | failed | in_progress; a run whose backend state this server does not " + "recognize reports that raw state verbatim, so treat any other value as unknown " + "rather than as a failure.",
     inputSchema: {
       workflowId: exports_external.string().describe("The workflow ID"),
       status: exports_external.enum(["success", "failed", "in_progress"]).optional().describe("Filter runs by outcome"),
@@ -53628,7 +53628,7 @@ function registerTools(server, ctx, capabilities) {
     });
   }));
   server.registerTool("run_workflow", {
-    description: "Run a workflow to extract fresh data. The run is asynchronous and may take several minutes. Do NOT poll or sleep-wait for completion. Return the workflow ID to the user and let them check status with get_workflow or fetch results later with fetch_data.",
+    description: "Run a workflow to extract fresh data, including retrying a failed one-time or scheduled workflow. Preserve the existing workflow ID and configuration - do not delete and recreate the workflow solely to retry it. The run is asynchronous and may take several minutes. Do NOT poll or sleep-wait for completion. Return the workflow ID to the user and let them check status with get_workflow or fetch results later with fetch_data. Realtime workflows cannot be manually run.",
     inputSchema: {
       workflowId: exports_external.string().describe("The workflow ID to run"),
       limit: exports_external.preprocess(coerceNumber(), exports_external.number()).optional().describe("Maximum number of records to extract (default: 1000)")
@@ -53651,7 +53651,7 @@ function registerTools(server, ctx, capabilities) {
   const FETCH_DATA_DEFAULT_LIMIT = 50;
   const FETCH_DATA_MAX_LIMIT = 500;
   server.registerTool("fetch_data", {
-    description: "Get a PAGE of extracted data from a workflow. Use ONLY for previews, sorted/filtered slices, or explicit 'first N rows' / 'top N' queries (capped at 500 rows per call). Do NOT use this to retrieve a full dataset, 'all rows', or anything the user wants to analyze in Excel / pandas / duckdb — use export_data for those. Data is only available after the workflow run has completed (status is no longer 'Running' or 'Validating'). Do NOT poll or sleep-wait for completion.",
+    description: "Get a PAGE of extracted data from a workflow. Use ONLY for previews, sorted/filtered slices, or explicit 'first N rows' / 'top N' queries (capped at 500 rows per call). Do NOT use this to retrieve a full dataset, 'all rows', or anything the user wants to analyze in Excel / pandas / duckdb - use export_data for those. Data is only available after the workflow run has completed (status is no longer 'Running' or 'Validating'). Do NOT poll or sleep-wait for completion.",
     inputSchema: {
       workflowId: exports_external.string().describe("The workflow ID"),
       limit: exports_external.preprocess(coerceNumber(), exports_external.number()).optional().describe(`Maximum number of records to return. Default ${FETCH_DATA_DEFAULT_LIMIT}, max ${FETCH_DATA_MAX_LIMIT}.`),
@@ -53686,7 +53686,7 @@ function registerTools(server, ctx, capabilities) {
     return jsonResult(result);
   }));
   server.registerTool("export_data", {
-    description: "PREFERRED tool for retrieving a workflow's FULL dataset. Materializes the data to object storage and returns a signed download URL. Use this whenever the user wants 'all rows', 'the full dataset', 'everything', an export, or anything destined for Excel / pandas / duckdb / a CSV file — even for small workflows. The URL is self-authenticating (open with `fetch(url)`, no Authorization header). Use fetch_data ONLY when the user explicitly asks for a small preview slice (e.g., 'first 10', 'top N sorted by X').",
+    description: "PREFERRED tool for retrieving a workflow's FULL dataset. Materializes the data to object storage and returns a signed download URL. Use this whenever the user wants 'all rows', 'the full dataset', 'everything', an export, or anything destined for Excel / pandas / duckdb / a CSV file - even for small workflows. The URL is self-authenticating (open with `fetch(url)`, no Authorization header). Use fetch_data ONLY when the user explicitly asks for a small preview slice (e.g., 'first 10', 'top N sorted by X').",
     inputSchema: {
       workflowId: exports_external.string().describe("The workflow ID"),
       format: exports_external.enum(["csv", "json"]).optional().describe("Export format. Default 'csv'."),
@@ -53812,7 +53812,7 @@ function registerTools(server, ctx, capabilities) {
     });
   }));
   server.registerTool("pause_workflow", {
-    description: "Pause an ACTIVE workflow so it stops running on its schedule. Requires the workflow to be in ACTIVE state — pausing a workflow that is currently running, already paused, or in PREVIEW will return an error. Use approve_workflow to resume a paused workflow.",
+    description: "Pause an ACTIVE workflow so it stops running on its schedule. Requires the workflow to be in ACTIVE state - pausing a workflow that is currently running, already paused, or in PREVIEW will return an error. Use approve_workflow to resume a paused workflow.",
     inputSchema: {
       workflowId: exports_external.string().min(1).describe("The workflow ID to pause")
     },
@@ -53826,17 +53826,17 @@ function registerTools(server, ctx, capabilities) {
     });
   }));
   server.registerTool("update_workflow", {
-    description: `Update a workflow's configuration. All fields are optional — only provided fields will be updated. Use this to change the name, URLs, extraction schema, entity, prompt, schedule, or other metadata.
+    description: `Update a workflow's configuration. All fields are optional - only provided fields will be updated. Use this to change the name, URLs, extraction schema, entity, prompt, schedule, or other metadata.
 
 ` + `IMPORTANT: You cannot change a workflow's interval to or from REAL_TIME. Julie realtime workflows are architecturally different from scheduled workflows and must be created with create_realtime_monitor from the start. Existing workflows cannot be converted between these modes in place; do not delete and recreate a workflow as an update workaround.
 
-` + "ASSISTANT-OWNED INTENT: Use request_workflow_update—not `userPrompt`—for agent-built workflow changes to extraction intent, navigation, pagination, data sourcing, repair, or generated scripts. `userPrompt` may be rejected with `SHELLY_INTENT_REQUIRES_EXTRACTION_SPEC` because those workflows' canonical intent is owned by the Assistant. " + "Call get_workflow first. If its template.controlledParts contains the setting, create and apply a template version instead of overriding the workflow directly. NEVER delete and recreate a workflow to work around an update limitation—that changes workflowId, breaks downstream tables/connectors, and discards history.",
+` + "ASSISTANT-OWNED INTENT: Use request_workflow_update - not `userPrompt` - for agent-built workflow changes to extraction intent, navigation, pagination, data sourcing, repair, or generated scripts. `userPrompt` may be rejected with `SHELLY_INTENT_REQUIRES_EXTRACTION_SPEC` because those workflows' canonical intent is owned by the Assistant. " + "Call get_workflow first. If its template.controlledParts contains the setting, create and apply a template version instead of overriding the workflow directly. NEVER delete and recreate a workflow to work around an update limitation - that changes workflowId, breaks downstream tables/connectors, and discards history.",
     inputSchema: strictSchema({
       workflowId: exports_external.string().describe("The workflow ID to update"),
       name: exports_external.string().optional().describe("New name for the workflow"),
       urls: exports_external.preprocess(coerceArray(true), exports_external.array(exports_external.string()).min(1)).optional().describe("New target URLs for the workflow (array of strings). Also accepts a single URL string."),
       entity: exports_external.string().optional().describe("Entity name for extraction (e.g., 'Product', 'Job Posting')"),
-      schema: exports_external.preprocess(coerceArray(), exports_external.array(exports_external.object(SchemaFieldShape).strict())).optional().describe("New extraction schema fields. This REPLACES the whole schema — list every field you want to keep, each with its isKey flag, or the omitted ones and their flags are lost. Call get_workflow first and edit the schema it returns."),
+      schema: exports_external.preprocess(coerceArray(), exports_external.array(exports_external.object(SchemaFieldShape).strict())).optional().describe("New extraction schema fields. This REPLACES the whole schema - list every field you want to keep, each with its isKey flag, or the omitted ones and their flags are lost. Call get_workflow first and edit the schema it returns."),
       description: exports_external.string().max(500).optional().describe("Workflow description (max 500 characters)"),
       tags: exports_external.preprocess(coerceArray(true), exports_external.array(exports_external.string())).optional().describe("Tags for organizing workflows"),
       userPrompt: exports_external.string().optional().describe("Navigation prompt for agentic-navigation mode (10-5000 characters)"),
@@ -53858,7 +53858,7 @@ function registerTools(server, ctx, capabilities) {
         "FOUR_WEEKS",
         "MONTHLY",
         "CUSTOM"
-      ]).optional().describe("How often the workflow should run. CUSTOM requires schedules. Note: REAL_TIME is NOT allowed here — realtime workflows must be created via create_realtime_monitor."),
+      ]).optional().describe("How often the workflow should run. CUSTOM requires schedules. Note: REAL_TIME is NOT allowed here - realtime workflows must be created via create_realtime_monitor."),
       schedules: exports_external.preprocess(coerceArray(true), exports_external.array(exports_external.string())).optional().describe("Cron expressions for CUSTOM update interval"),
       timezone: exports_external.string().min(1).optional().describe("IANA timezone for cron schedules, for example America/New_York. A timezone-only update preserves existing cron expressions."),
       location: exports_external.preprocess(coerceJson(), LocationSchema).optional().describe("Scraping location: {type:'auto'} or {type:'manual', isoCode:'US'}"),
@@ -54277,22 +54277,29 @@ function registerTools(server, ctx, capabilities) {
     });
   }));
   server.registerTool("list_templates", {
-    description: "List all templates in the current team. Templates define reusable configurations (prompt, schema, notifications).",
+    description: "List all templates in the current team. Templates define reusable configurations (prompt, schema, validation rules, notifications, and frequency). Use this to find a matching template before creating a workflow. To instantiate a template, call create_workflow with its templateId, source URLs, and optional templateVersion - never copy the template prompt, entity, or schema into standalone creation.",
     inputSchema: strictSchema({}),
     annotations: { readOnlyHint: true, destructiveHint: false, idempotentHint: true }
   }, withErrorHandling("list_templates", async () => {
     const templates = await ctx.client.template.list();
-    return jsonResult({ templates, count: templates.length });
+    return jsonResult({
+      templates,
+      count: templates.length,
+      instructions: "To instantiate a listed template, call create_workflow with the matching templateId, source URLs, and optional templateVersion. Do not copy the template prompt, entity, or schema into standalone creation."
+    });
   }));
   server.registerTool("get_template", {
-    description: "Get a template by ID, including all published versions.",
+    description: "Get a template by ID, including all published versions and their schemas. After inspecting it, instantiate the template with create_workflow using templateId, source URLs, and optional templateVersion. Never copy the returned prompt, entity, or schema into standalone creation - those values are inherited when templateId is passed.",
     inputSchema: strictSchema({
       templateId: exports_external.string().describe("The template ID")
     }),
     annotations: { readOnlyHint: true, destructiveHint: false, idempotentHint: true }
   }, withErrorHandling("get_template", async (args) => {
     const template = await ctx.client.template.get(args.templateId);
-    return jsonResult({ template });
+    return jsonResult({
+      template,
+      instructions: `To instantiate this template, call create_workflow with templateId "${args.templateId}", source URLs, and optional templateVersion. Do not copy this template's prompt, entity, or schema into standalone creation.`
+    });
   }));
   server.registerTool("create_template", {
     description: "Create a new template. After creation, use create_template_version to publish a version with prompt, schema, and notifications.",
@@ -54315,7 +54322,7 @@ function registerTools(server, ctx, capabilities) {
   server.registerTool("update_template", {
     description: `Update a template's name or description ONLY. At least one of the two must be provided.
 
-` + "This tool CANNOT change what linked workflows inherit — prompt, schema, validation rules, notifications and frequency all live in template *versions*, which are immutable snapshots. " + "To change any of those, publish a new version with `create_template_version` (the version IS the edit), then roll it out to linked workflows with `apply_template_version`. " + "There is no in-place edit of a published version, and never report a prompt or schema change as impossible — versioning is the supported path.",
+` + "This tool CANNOT change what linked workflows inherit - prompt, schema, validation rules, notifications and frequency all live in template *versions*, which are immutable snapshots. " + "To change any of those, publish a new version with `create_template_version` (the version IS the edit), then roll it out to linked workflows with `apply_template_version`. " + "There is no in-place edit of a published version, and never report a prompt or schema change as impossible - versioning is the supported path.",
     inputSchema: strictSchema({
       templateId: exports_external.string().describe("The template ID to update"),
       name: exports_external.preprocess(coerceNull(), exports_external.string().optional()).optional().describe("New template name"),
@@ -54371,7 +54378,7 @@ function registerTools(server, ctx, capabilities) {
     }))).optional().describe("Predefined categories for a CLASSIFICATION field ({title, definition}[]). Required for fieldType=CLASSIFICATION; omitted otherwise.")
   };
   server.registerTool("create_template_version", {
-    description: `Publish a new version of a template. Versions capture the full workflow config: prompt, schema, and notifications. All fields are optional — include only what this version should set.
+    description: `Publish a new version of a template. Versions capture the full workflow config: prompt, schema, and notifications. All fields are optional - include only what this version should set.
 
 ` + "THIS IS HOW YOU EDIT A TEMPLATE'S PROMPT OR SCHEMA. Published versions are immutable, so 'changing the template prompt' means publishing a new version here and then calling `apply_template_version` to push it onto linked workflows. " + "`update_template` only renames a template; it cannot touch prompt or schema.",
     inputSchema: strictSchema({
@@ -54380,7 +54387,7 @@ function registerTools(server, ctx, capabilities) {
       schemaId: exports_external.string().optional().describe("Existing schema ID to reference (mutually exclusive with schemaFields)"),
       schemaFields: exports_external.preprocess(coerceArray(), exports_external.array(exports_external.object(TemplateSchemaFieldShape).strict())).optional().describe("Inline schema fields to create a new schema (mutually exclusive with schemaId)"),
       schemaEntity: exports_external.string().optional().describe("Entity name for the inline schema"),
-      schemaValidationRules: exports_external.preprocess(coerceJson(), exports_external.record(exports_external.string(), exports_external.record(exports_external.string(), exports_external.unknown()))).optional().describe("Per-field schema validation rules, keyed by field name. Not inherited from prior versions — omitting this on a new version drops any rules the previous version had."),
+      schemaValidationRules: exports_external.preprocess(coerceJson(), SchemaValidationRulesSchema).optional().describe('Per-field schema validation rules keyed by field name. Each rule uses kind STRING, NUMBER, OTHER, OBJECT, or ARRAY and nested rules are supported. Rule metadata requires editedBy (default|llm|agent|ops|user) and an ISO-8601 editedAt timestamp. Example: { "price": { "kind": "NUMBER", "minimum": { "value": 0, "editedBy": "user", "editedAt": "2026-01-01T00:00:00.000Z" } } }. Not inherited from prior versions - omitting this on a new version drops any rules the previous version had.'),
       notifications: exports_external.preprocess(coerceArray(), exports_external.array(exports_external.object({
         eventType: exports_external.string().describe("Notification event type"),
         eventConfiguration: exports_external.preprocess(coerceJson(), exports_external.record(exports_external.string(), exports_external.unknown())).optional(),
@@ -54461,7 +54468,7 @@ function registerTools(server, ctx, capabilities) {
     return jsonResult({ schemas: schemas4, count: schemas4.length });
   }));
   server.registerTool("link_template_to_workflows", {
-    description: "Link one or more EXISTING workflows to a template in a single call. " + "This is the bulk equivalent of creating a workflow with `templateId`: linked workflows adopt the template's configuration (prompt, schema, notifications) and stay in sync with it. " + "The template ENFORCES its schema on linked workflows — their extracted output conforms to the template's declared field names, so this is the way to make many workflows produce a consistent, canonical schema. " + "Use `list_templates`/`get_template` to find the template, and `list_workflows` to find the workflow IDs. " + "Set `force: true` to relink workflows already linked to a different template.",
+    description: "Link one or more EXISTING workflows to a template in a single call. " + "This is the bulk equivalent of creating a workflow with `templateId`: linked workflows adopt the template's configuration (prompt, schema, notifications) and stay in sync with it. " + "The template ENFORCES its schema on linked workflows - their extracted output conforms to the template's declared field names, so this is the way to make many workflows produce a consistent, canonical schema. " + "Use `list_templates`/`get_template` to find the template, and `list_workflows` to find the workflow IDs. " + "Set `force: true` to relink workflows already linked to a different template.",
     inputSchema: strictSchema({
       templateId: exports_external.string().describe("The template ID to link workflows to"),
       workflowIds: exports_external.preprocess(coerceArray(true), exports_external.array(exports_external.string()).min(1)).describe("Workflow IDs to link to the template (array of strings). Also accepts a single ID string."),
@@ -54478,7 +54485,7 @@ function registerTools(server, ctx, capabilities) {
       return jsonResult({
         success: false,
         conflicts: result.conflicts,
-        message: `Cannot link — ${result.conflicts.length} workflow(s) are already linked to another template: ${summary}. ` + "Re-run with force: true to move them to this template."
+        message: `Cannot link - ${result.conflicts.length} workflow(s) are already linked to another template: ${summary}. ` + "Re-run with force: true to move them to this template."
       });
     }
     return jsonResult({
@@ -54538,7 +54545,7 @@ function registerTools(server, ctx, capabilities) {
     });
   }));
 }
-var SchemaFieldShape, LocationSchema, MonitoringValueOperators, MonitoringValuelessOperators, MonitoringConditionOperatorSchema, MonitoringConditionSchema, MonitoringSchema, RESUMABLE_ASSISTANT_STATUSES, ACTIVE_ASSISTANT_STATUSES, IDLE_ASSISTANT_STATUSES, CLOSED_ASSISTANT_STATUSES, DASHBOARD_BASE_URL = "https://www.kadoa.com", WORKFLOW_AUDIT_WATCHED_KEYS;
+var SchemaFieldShape, SchemaValidationAttributionShape, SchemaValidationPresenceRule, SchemaValidationUniquenessRule, SchemaValidationStringLengthRule, SchemaValidationStringFormatRule, SchemaValidationFieldRulesSchema, SchemaValidationRulesSchema, LocationSchema, MonitoringValueOperators, MonitoringValuelessOperators, MonitoringConditionOperatorSchema, MonitoringConditionSchema, MonitoringSchema, RESUMABLE_ASSISTANT_STATUSES, ACTIVE_ASSISTANT_STATUSES, IDLE_ASSISTANT_STATUSES, CLOSED_ASSISTANT_STATUSES, DASHBOARD_BASE_URL = "https://www.kadoa.com", WORKFLOW_AUDIT_WATCHED_KEYS;
 var init_tools = __esm(() => {
   init_dist2();
   init_zod();
@@ -54548,8 +54555,105 @@ var init_tools = __esm(() => {
     description: exports_external.string().optional().describe("What this field contains"),
     example: exports_external.string().describe("Example value"),
     dataType: exports_external.enum(["STRING", "NUMBER", "BOOLEAN", "DATE", "DATETIME", "MONEY", "IMAGE", "LINK", "OBJECT", "ARRAY"]).optional().describe("Data type for the field"),
-    isKey: exports_external.preprocess(coerceBoolean(), exports_external.boolean()).optional().describe("Marks this field as a key field — the stable identity used to match records across runs. Change detection diffs records by their key fields, so a monitored or real-time workflow without one cannot tell an updated record from a new one. Set it on whatever uniquely identifies a row (a detail URL, an ID, a ticker).")
+    isKey: exports_external.preprocess(coerceBoolean(), exports_external.boolean()).optional().describe("Marks this field as a key field - the stable identity used to match records across runs. Change detection diffs records by their key fields, so a monitored or real-time workflow without one cannot tell an updated record from a new one. Set it on whatever uniquely identifies a row (a detail URL, an ID, a ticker).")
   };
+  SchemaValidationAttributionShape = {
+    editedBy: exports_external.enum(["default", "llm", "agent", "ops", "user"]).describe("Who last edited this rule: default, llm, agent, ops, or user"),
+    editedByLabel: exports_external.string().optional().describe("Optional editor identifier"),
+    editedAt: exports_external.string().datetime().describe("ISO-8601 timestamp when this rule was last edited")
+  };
+  SchemaValidationPresenceRule = exports_external.object({
+    target: exports_external.number().int().min(0).max(100).step(20).describe("Expected percentage of rows with a value"),
+    ...SchemaValidationAttributionShape
+  }).strict();
+  SchemaValidationUniquenessRule = exports_external.object({
+    target: exports_external.number().int().min(0).max(100).step(20).describe("Expected percentage of rows with a unique value"),
+    ...SchemaValidationAttributionShape
+  }).strict();
+  SchemaValidationStringLengthRule = exports_external.object({
+    value: exports_external.number().int().min(0).describe("String length bound"),
+    ...SchemaValidationAttributionShape
+  }).strict();
+  SchemaValidationStringFormatRule = exports_external.discriminatedUnion("kind", [
+    exports_external.object({
+      kind: exports_external.literal("FREE_TEXT"),
+      charset: exports_external.discriminatedUnion("kind", [
+        exports_external.object({
+          kind: exports_external.literal("PRESET"),
+          preset: exports_external.enum(["natural_language", "alphanumeric", "alpha"])
+        }).strict()
+      ]),
+      ...SchemaValidationAttributionShape
+    }).strict(),
+    exports_external.object({
+      kind: exports_external.literal("FORMAT"),
+      source: exports_external.discriminatedUnion("kind", [
+        exports_external.object({
+          kind: exports_external.literal("PRESET"),
+          preset: exports_external.enum(["url", "email", "phone", "date", "datetime", "time", "uuid", "slug"])
+        }).strict(),
+        exports_external.object({
+          kind: exports_external.literal("CUSTOM"),
+          pattern: exports_external.string().min(1).describe("Regular expression pattern")
+        }).strict()
+      ]),
+      ...SchemaValidationAttributionShape
+    }).strict(),
+    exports_external.object({
+      kind: exports_external.literal("LIST"),
+      source: exports_external.discriminatedUnion("kind", [
+        exports_external.object({
+          kind: exports_external.literal("PRESET"),
+          preset: exports_external.enum(["language2", "country2", "country3", "currency3", "month3", "usState2"])
+        }).strict(),
+        exports_external.object({
+          kind: exports_external.literal("CUSTOM"),
+          values: exports_external.array(exports_external.string().min(1)).min(1).describe("Allowed string values")
+        }).strict()
+      ]),
+      ...SchemaValidationAttributionShape
+    }).strict()
+  ]);
+  SchemaValidationFieldRulesSchema = exports_external.lazy(() => exports_external.discriminatedUnion("kind", [
+    exports_external.object({
+      kind: exports_external.literal("STRING"),
+      presence: SchemaValidationPresenceRule.optional(),
+      uniqueness: SchemaValidationUniquenessRule.optional(),
+      minLength: SchemaValidationStringLengthRule.optional(),
+      maxLength: SchemaValidationStringLengthRule.optional(),
+      minHtmlElements: SchemaValidationStringLengthRule.optional(),
+      maxHtmlElements: SchemaValidationStringLengthRule.optional(),
+      format: SchemaValidationStringFormatRule.optional()
+    }).strict(),
+    exports_external.object({
+      kind: exports_external.literal("NUMBER"),
+      presence: SchemaValidationPresenceRule.optional(),
+      uniqueness: SchemaValidationUniquenessRule.optional(),
+      minimum: exports_external.object({ value: exports_external.number().describe("Minimum numeric value"), ...SchemaValidationAttributionShape }).strict().optional(),
+      maximum: exports_external.object({ value: exports_external.number().describe("Maximum numeric value"), ...SchemaValidationAttributionShape }).strict().optional(),
+      maxDecimalPlaces: exports_external.object({ value: exports_external.number().int().min(0).max(16), ...SchemaValidationAttributionShape }).strict().optional()
+    }).strict(),
+    exports_external.object({
+      kind: exports_external.literal("OTHER"),
+      presence: SchemaValidationPresenceRule.optional(),
+      uniqueness: SchemaValidationUniquenessRule.optional()
+    }).strict(),
+    exports_external.object({
+      kind: exports_external.literal("OBJECT"),
+      presence: SchemaValidationPresenceRule.optional(),
+      uniqueness: SchemaValidationUniquenessRule.optional(),
+      properties: exports_external.record(exports_external.string(), SchemaValidationFieldRulesSchema)
+    }).strict(),
+    exports_external.object({
+      kind: exports_external.literal("ARRAY"),
+      presence: SchemaValidationPresenceRule.optional(),
+      uniqueness: SchemaValidationUniquenessRule.optional(),
+      minItems: exports_external.object({ value: exports_external.number().int().min(0), ...SchemaValidationAttributionShape }).strict().optional(),
+      maxItems: exports_external.object({ value: exports_external.number().int().min(0), ...SchemaValidationAttributionShape }).strict().optional(),
+      items: SchemaValidationFieldRulesSchema.optional()
+    }).strict()
+  ]));
+  SchemaValidationRulesSchema = exports_external.record(exports_external.string(), SchemaValidationFieldRulesSchema);
   LocationSchema = exports_external.object({
     type: exports_external.enum(["auto", "manual"]),
     isoCode: exports_external.string().trim().min(2).optional()
@@ -54638,7 +54742,7 @@ var package_default;
 var init_package = __esm(() => {
   package_default = {
     name: "@kadoa/mcp",
-    version: "0.5.20",
+    version: "0.5.21",
     description: "Kadoa MCP Server — manage workflows from Claude Desktop, Cursor, and other MCP clients",
     type: "module",
     main: "dist/index.js",
@@ -60241,21 +60345,22 @@ async function createServer(auth, options) {
   const server = new McpServer({ name: "kadoa", version: package_default.version }, {
     instructions: [
       "IMPORTANT: One workflow = one source. A single workflow can extract many different fields, tables, and sections from the same URL(s) using a rich schema.",
-      "Do NOT create multiple workflows for different data points on the same page \u2014 instead, define one workflow with a multi-field schema covering everything needed.",
+      "Do NOT create multiple workflows for different data points on the same page - instead, define one workflow with a multi-field schema covering everything needed.",
       "",
       "Kadoa workflows use agentic navigation: the AI agent can browse pages, click buttons, fill forms, select dropdowns, paginate, open detail pages, and handle multi-step interactions.",
       "Describe the full navigation steps in the prompt (e.g., 'select year 2022 from the dropdown, click Search, extract the table, then repeat for 2023').",
       "",
-      "Workflow lifecycle: create_workflow \u2192 get_workflow (check status) \u2192 fetch_data (get results). Workflows run asynchronously \u2014 never poll or sleep-wait.",
+      "Workflow lifecycle: create_workflow \u2192 get_workflow (check status) \u2192 fetch_data (get results). Workflows run asynchronously - never poll or sleep-wait.",
       "",
       "Use create_realtime_monitor only when the user wants continuous Julie change detection with alerts. It first persists notification channels, then asynchronously returns workflow/session/thread/job IDs; use the workflow Assistant tools for follow-up status, questions, and controls.",
       "For one-time or scheduled extraction, use create_workflow. Use scrape for an immediate raw HTML or markdown fetch from one URL. Use create_workflow for structured extraction, recurring runs, monitoring, or navigation-heavy jobs.",
+      "If a one-time or scheduled workflow fails, retry it with run_workflow using the existing workflow ID. Do NOT delete and recreate the workflow just to retry it - preserve its workflow ID and configuration. Realtime workflows cannot be manually run.",
       "Use list_changes and get_change to retrieve detected diffs from realtime monitoring workflows.",
       "",
       "Schema tips: Use descriptive field names and examples. Group related data under one entity.",
-      "The AI agent uses the schema + prompt to understand what to extract \u2014 a detailed prompt with a comprehensive schema produces better results than multiple simple workflows.",
+      "The AI agent uses the schema + prompt to understand what to extract - a detailed prompt with a comprehensive schema produces better results than multiple simple workflows.",
       "",
-      "Templates enforce their schema: a workflow created from a template (create_workflow with templateId) or linked to one (link_template_to_workflows) produces output whose field names conform to the template's declared schema. Use templates when you need many workflows to return a consistent, canonical set of fields \u2014 do NOT assume the extractor will drift field names away from a template's schema."
+      "Templates enforce their schema: a workflow created from a template (create_workflow with templateId) or linked to one (link_template_to_workflows) produces output whose field names conform to the template's declared schema. Use templates when you need many workflows to return a consistent, canonical set of fields - do NOT assume the extractor will drift field names away from a template's schema."
     ].join(`
 `)
   });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@kadoa/mcp",
-  "version": "0.5.20",
+  "version": "0.5.21",
   "description": "Kadoa MCP Server — manage workflows from Claude Desktop, Cursor, and other MCP clients",
   "type": "module",
   "main": "dist/index.js",