project-graph-mcp 1.3.0 → 2.0.0

package/README.md

@@ -1,14 +1,13 @@
 [![npm version](https://img.shields.io/npm/v/project-graph-mcp)](https://www.npmjs.com/package/project-graph-mcp)
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 [![Node.js](https://img.shields.io/badge/Node.js-%3E%3D18-339933?logo=node.js&logoColor=white)](https://nodejs.org)
-[![Zero Dependencies](https://img.shields.io/badge/dependencies-0-brightgreen)](https://www.npmjs.com/package/project-graph-mcp)
 
 # project-graph-mcp
 
-An MCP server that parses your source code into a **10-50x compressed skeleton** — classes, functions, imports, and dependencies in a minified JSON. Agents navigate the graph using `expand`, `deps`, and `usages` without reading irrelevant files.
+An MCP server that parses your source code into a **10-50x compressed skeleton** — classes, functions, imports, and dependencies in a minified JSON. Agents navigate the graph using `expand`, `deps`, and `usages` without reading irrelevant files. The **AI Context Layer** compresses an entire codebase into ~1700 tokens (97% savings) with a single `get_ai_context` call. Supports **monorepo scanning** and **streaming analysis** for large codebases.
 
 > [!TIP]
-> **132 kB, 47 files, zero external dependencies.** Add one line to your MCP config and the server downloads itself on the next IDE restart.
+> **18 MCP tools.** Add one line to your MCP config and the server downloads itself on the next IDE restart.
 
 ### Project Skeleton (10-50x compression)
 
@@ -44,27 +43,213 @@ stripStringsAndComments(code, { backtick: true, templateInterpolation: false })
 - **Duplicate detection** — finds functionally similar functions by signature + structure similarity
 - **Large file analysis** — candidates for splitting by lines, functions, and exports count
 - **Legacy pattern finder** — outdated code patterns and redundant npm deps (built into Node.js 18+)
-- **Health Score (0-100)** — aggregated result from all checks in one call
+- **JSDoc consistency** — validates `@param` count/names and `@returns` against AST signatures
+- **Type checking** — optional `tsc --checkJs` wrapper with graceful fallback
+- **Health Score (0-100)** — aggregated result from all checks in one call (`get_full_analysis` or quick `get_analysis_summary`)
+- **Incremental cache** — per-file analysis results cached in `.context/.cache/` with content hashing
+
+### AI Context Layer
+
+One call loads everything an agent needs to understand a project:
+
+```javascript
+get_ai_context({ path: "src/" })
+// → { skeleton, docs, totalTokens: 1742, savings: "97%" }
+```
+
+- **Code compression** — Terser-minified source with export legend headers (20-55% per file)
+- **Compact Code Mode** — project-wide `compact`/`beautify` (preserves all names, strips comments/whitespace)
+- **Doc Dialect** — compact `.context/` documentation format, auto-generated from AST with `{DESCRIBE}` markers. Descriptions use a **simplified English dialect** — max 80 characters, pipe-separated, with standard abbreviations (`fn/ret/cfg/init/auth/db/msg`):
+
+```
+generateJSDoc(t)→e.push,e.join|build /** */ JSDoc block from .ctx signature
+export expandFile(e,n)→beautify,inject|beautify compact JS + inject JSDoc from .ctx
+PATTERNS: Terser beautify|AST walk for injection points|reverse-order insertion
+```
+- **Two-tier `.ctx`** — `.ctx` (machine-generated, AST signatures) + `.ctx.md` (agent notes, TODO, decisions)
+- **Self-enriching** — `@enrich` instructions embedded in `.ctx` files guide any AI agent to fill descriptions
+- **Staleness detection** — `@sig` hashes track structural changes; `check_stale_docs` identifies outdated docs
+- **Merge strategy** — regenerating `.ctx` files preserves existing descriptions
+- **Boot aggregator** — `get_ai_context` combines skeleton + docs + compressed files in one response
+
+```bash
+# Generate .context/ documentation templates
+npx project-graph-mcp generate-ctx src/
+
+# View project docs in compact format
+npx project-graph-mcp docs src/
+
+# Compress a single file for AI
+npx project-graph-mcp compress src/core/parser.js
+```
+
+### Two-Tier Context: Overview → Focus
+
+Agents don't need full context for every file. The server provides a **progressive loading model**:
+
+| Mode | What agent reads | Token cost | Use case |
+|------|-----------------|------------|----------|
+| **Overview** | `get_skeleton()` + compact code | codeTok only | Understand project structure |
+| **Focus** | compact code + `.ctx` for specific files | codeTok + ctxTok (per file) | Deep work on area of interest |
+| **Traditional** | All raw source files | expanded | Reading source directly |
+
+```javascript
+// 1. Overview: read entire project structure (cheap, no .ctx)
+get_skeleton({ path: "src/" })
+// → Legend, stats, all classes/functions/exports — ~2-5K tokens
+
+// 2. Focus: get enriched context ONLY for area of interest
+get_focus_zone({ recentFiles: ["src/core/parser.js", "src/mcp/tools.js"] })
+// → Compact code + .ctx documentation for just those 2 files
+
+// 3. Or auto-detect from git diff
+get_focus_zone({ path: ".", useGitDiff: true })
+// → Context for recently changed files only
+```
+
+**Real-world token budget** (45-file project):
+
+```
+Traditional (raw sources):     64K tok — read everything expanded
+Overview (compact, no .ctx):   47K tok — 27% savings, full project
+Focus (compact + 3 files ctx): 47.6K tok — same savings + semantic descriptions
+```
+
+Compact code saves **27% on average** (34-38% for typical files) by stripping whitespace and comments while preserving all names. The `.ctx` cost is **per-file** (~200 tok each) and only loaded for the focus area — the agent gets typed signatures and descriptions that the raw source never had.
+
+**Per-file metrics breakdown:**
+
+| Layer | What | Tokens |
+|-------|------|--------|
+| Code (compact .js) | Minified source, all names preserved | codeTok |
+| Context (.ctx) | AST signatures, types, descriptions | ctxTok |
+| Total (focus mode) | What agent reads when focusing | codeTok + ctxTok |
+| Expanded (beautified) | What a human would read (raw source) | expanded |
+
+### Compact Code Architecture
+
+Three modes for AI-native codebase editing — configure per project via `.context/config.json`:
+
+| Mode | Storage | Agent reads | Agent edits |
+|------|---------|-------------|-------------|
+| **1 — Native Compact** | Minified JS | Files directly | Files directly |
+| **2 — Full Storage** (default) | Formatted JS | `get_compressed_file` | `edit_compressed` |
+| **3 — IDE Virtual** | Minified JS | IDE renders full view | IDE handles mapping |
+
+**Mode 2 workflow** (recommended):
+
+```javascript
+// 1. Read compressed view (saves 20-55% tokens)
+get_compressed_file({ path: "src/parser.js" })
+
+// 2. Edit by symbol name — server finds it via AST
+edit_compressed({
+  path: "src/parser.js",
+  symbol: "parseFile",
+  code: "export async function parseFile(code, filename) { /* new body */ }"
+})
+
+// 3. Validate .ctx contracts after editing
+validate_ctx_contracts({ path: "." })
+```
+
+**`.ctx` typed signatures** — JSDoc types extracted into compact format:
+
+```
+parseFile(code:string,filename:string)→Promise<ParseResult>→parse,walk|parse JS file into AST
+```
+
+```bash
+# Set project mode
+npx project-graph-mcp set-mode . 2
+
+# Validate .ctx documentation matches source
+npx project-graph-mcp validate-ctx . --strict
+```
+
+### Expand & Validate Pipeline
+
+The compact→expand pipeline is **fully reversible**. Verify round-trip integrity:
+
+```javascript
+// Expand a compact file back to full formatted + JSDoc
+compact({ action: "expand_file", path: "src/parser.js" })
+
+// Expand entire project
+compact({ action: "expand_project", path: ".", dryRun: true })
+
+// Validate compact ↔ expand round-trip
+compact({ action: "validate_pipeline", path: ".", strict: true })
+// → Reports any functions in source missing from .ctx
+```
 
 ### Test Checklists
 
-JSDoc annotations (`@test` and `@expect`) define test checklists directly in the code:
+Test checklists live in `.ctx.md` files (alongside documentation), not in source code:
+
+```markdown
+## Tests
+- [ ] POST /api/users with valid data → 201 Created
+- [ ] GET /api/users/:id returns user object
+- [x] DELETE /api/users/:id → 204 No Content
+```
+
+The agent calls `get_pending_tests`, runs the test, then `mark_test_passed` (which writes `[x]` directly to the `.ctx.md` file). Test state is persistent and survives session restarts.
+
+### Monorepo Support
+
+`discover_sub_projects` scans standard monorepo directories (`packages/`, `apps/`, `services/`, `modules/`, `libs/`, `plugins/`) for sub-projects with `package.json`. Combined with `parseProject({ recursive: true })`, agents can analyze entire monorepos.
+
+### Database Analysis
+
+Scan SQL migrations and code for database schema insights:
 
 ```javascript
-/**
- * Create new user via API
- *
- * @test request: POST /api/users with valid data
- * @expect status: 201 Created
- */
-async createUser(data) { ... }
+// Extract schema from SQL/migration files
+db({ action: "schema", path: "src/" })
+
+// Find where each table is referenced in code
+db({ action: "table_usage", path: "src/", table: "users" })
+
+// Detect tables defined but never queried
+db({ action: "dead_tables", path: "src/" })
 ```
 
-The agent calls `get_pending_tests`, runs the test, then `mark_test_passed`. Supports browser, API, CLI, and integration test types.
+### Web Dashboard
+
+Every project-graph-mcp instance includes a built-in web UI at `http://localhost:{port}/`:
+
+- **Multi-project dashboard** — overview of all registered projects with token metrics
+- **File tree** — navigate project structure
+- **Code viewer** — compact/raw toggle with syntax highlighting and per-file compression stats
+- **Dependency graph** — visual dependency exploration
+- **Health panel** — analysis results
+- **Live monitor** — real-time agent activity via WebSocket
+
+With the optional gateway, all projects are accessible under `http://project-graph.local/{project-name}/`.
+
+### Compression Metrics
+
+Token-level metrics are available project-wide and per-file:
+
+```javascript
+// Project-wide: how many tokens for the entire codebase
+// GET /api/compression-stats
+// → { files: 45, codeTok: 47000, ctxTok: 9500, totalTok: 56500 }
+
+// Per-file: shown in code viewer header
+// 2054 + 527 ctx = 2581 → 3340 tok (23% savings)
+```
+
+### Performance
+
+- **Batch concurrency** — `generate_context_docs` processes 5 files in parallel
+- **Quick health check** — `get_analysis_summary` runs only cached per-file metrics (skips expensive cross-file analysis)
+- **Streaming analysis** — `getFullAnalysisStreaming` yields results incrementally as each sub-analysis completes
 
 ### Custom Rules & Framework References
 
-10 pre-built rulesets (62 rules) for React 18/19, Vue 3, Next.js 15, Express 5, Fastify 5, NestJS 10, TypeScript 5, Node.js 22, and [Symbiote.js](https://github.com/symbiotejs/symbiote.js). The server auto-detects your project type and returns adapted documentation via `get_framework_reference`.
+11 pre-built rulesets (86 rules) for React 18/19, Vue 3, Next.js 15, Express 5, Fastify 5, NestJS 10, TypeScript 5, Node.js 22, and [Symbiote.js](https://github.com/symbiotejs/symbiote.js). The server auto-detects your project type and returns adapted documentation via `get_framework_reference`.
 
 Custom project conventions can be added in the `rules/` directory or configured by the agent via `set_custom_rule`.
 
@@ -93,6 +278,19 @@ Add to your IDE's MCP configuration:
 
 Restart your IDE — project-graph-mcp will be downloaded and started automatically.
 
+#### Grouped Tools (v2.0)
+
+v2.0 uses 18 domain-grouped tools instead of 49 individual endpoints. Grouped tools use an `action` parameter:
+
+```javascript
+navigate({ action: "expand", symbol: "MyClass" })
+analyze({ action: "complexity", path: "src/" })
+docs({ action: "generate", path: ".", scope: "focus" })
+compact({ action: "compact_file", path: "src/parser.js" })
+```
+
+10 standalone tools (`get_skeleton`, `get_ai_context`, `invalidate_cache`, etc.) remain unchanged.
+
 <details>
 <summary>Where is my MCP config file?</summary>
 
@@ -115,7 +313,7 @@ See **[CONFIGURATION.md](CONFIGURATION.md)** for all supported IDEs (Antigravity
 git clone https://github.com/rnd-pro/project-graph-mcp
 cd project-graph-mcp
 # No npm install needed — zero dependencies
-# Use "node /path/to/project-graph-mcp/src/server.js" as the command in MCP config
+# Use "node /path/to/project-graph-mcp/src/network/server.js" as the command in MCP config
 ```
 
 </details>
@@ -130,6 +328,12 @@ npx project-graph-mcp deadcode src/       # Find unused code
 npx project-graph-mcp complexity src/     # Cyclomatic complexity
 npx project-graph-mcp similar src/        # Find duplicates
 npx project-graph-mcp pending src/        # List pending tests
+npx project-graph-mcp compress src/f.js   # Compress file for AI
+npx project-graph-mcp docs src/           # Project docs (doc-dialect)
+npx project-graph-mcp generate-ctx src/   # Generate .context/ docs
+npx project-graph-mcp validate-ctx .      # Validate .ctx ↔ source
+npx project-graph-mcp mode .             # Show current editing mode
+npx project-graph-mcp set-mode . 2       # Set mode (1/2/3)
 npx project-graph-mcp help                # All commands
 ```
 
@@ -164,8 +368,10 @@ Best used together with [**agent-pool-mcp**](https://www.npmjs.com/package/agent
 
 - [CONFIGURATION.md](CONFIGURATION.md) — Setup for all supported IDEs
 - [ARCHITECTURE.md](ARCHITECTURE.md) — Source code structure
-- [AGENT_ROLE.md](AGENT_ROLE.md) — Full system prompt for agents
-- [AGENT_ROLE_MINIMAL.md](AGENT_ROLE_MINIMAL.md) — Minimal variant (agent self-discovers)
+- [AGENT_ROLE.md](docs/examples/AGENT_ROLE.md) — Full system prompt for agents
+- [AGENT_ROLE_MINIMAL.md](docs/examples/AGENT_ROLE_MINIMAL.md) — Minimal variant (agent self-discovers)
+- [GUIDE.md](GUIDE.md) — Comprehensive usage guide with all tools
+- [ROADMAP.md](docs/ROADMAP.md) — Feature roadmap and backlog
 
 ## Related Projects
 - [agent-pool-mcp](https://github.com/rnd-pro/agent-pool-mcp) — Multi-agent orchestration via Gemini CLI

package/{AGENT_ROLE.md → docs/examples/AGENT_ROLE.md}

@@ -5,27 +5,37 @@ You have access to **Project Graph MCP** — a suite of code analysis and projec
 ## 🧭 Navigation & Understanding
 | Tool | Purpose |
 |------|---------|
-| `get_structure` | Get file/folder tree |
-| `get_skeleton` | Get code structure (classes, functions, exports) |
+| `get_skeleton` | Get compact code structure (classes, functions, exports) |
 | `expand` | Deep dive into a class or function |
+| `get_focus_zone` | Get enriched context for recently modified files |
+| `get_call_chain` | Find shortest path between two symbols |
+| `usages` | Find all usages of a symbol across the project |
+| `deps` | Get dependency tree for a symbol |
 | `get_agent_instructions` | Get project coding guidelines |
 | `get_framework_reference` | Get framework AI reference (auto-detects or explicit) |
+| `get_usage_guide` | Get full usage guide with examples |
+| `invalidate_cache` | Refresh graph after code changes (MANDATORY after edits) |
 
 ## 🧪 Testing System
 
 | Tool | Purpose |
 |------|---------|
-| `get_pending_tests` | List @test/@expect annotations needing verification |
-| `mark_test_passed` / `mark_test_failed` | Track test results |
+| `get_pending_tests` | List `[ ]` checklists from `.ctx.md` files |
+| `mark_test_passed` / `mark_test_failed` | Write `[x]` or `[!]` directly to `.ctx.md` |
 | `get_test_summary` | Progress report |
+| `reset_test_state` | Reset all checklists to `[ ]` |
 
-### When to Write @test/@expect
-Add annotations to JSDoc when creating or modifying **interactive methods**:
-- `onclick` / `onchange` / `oninput` event handlers
-- Methods that change DOM state (show/hide, toggle classes/attributes)
-- Navigation and routing methods
-- Form submission and validation handlers
-- Any method with user-visible side effects
+### How Test Checklists Work
+Tests live in `.ctx.md` files (the "agent zone" of the two-tier documentation), not in source code:
+
+```markdown
+# parser.js
+
+## Tests
+- [ ] Parse valid JS file with classes and functions
+- [ ] Handle syntax errors gracefully
+- [x] Parse empty file without crash
+```
 
 ### Browser Testing Workflow (VERIFICATION mode)
 After code changes, you MUST verify UI with this flow:
@@ -37,20 +47,9 @@ After code changes, you MUST verify UI with this flow:
 4. get_test_summary(path)            → final report before completing task
 ```
 
-**Rule**: If `get_pending_tests()` returns items, they MUST be executed in the browser before the task is marked complete. Never skip browser verification when @test annotations exist.
-
-### Example
-```javascript
-/**
- * Delete selected persona
- *
- * @test click: Click delete button on persona card
- * @test click: Confirm in dialog
- * @expect element: Persona removed from list
- * @expect visual: Toast notification appears
- */
-async onDeletePersona() { ... }
-```
+**Rule**: If `get_pending_tests()` returns items, they MUST be executed before the task is marked complete.
+
+> **Note**: Test state is persistent (written to files) and survives agent session restarts.
 
 ## 🗄️ Database Analysis
 | Tool | Purpose |
@@ -71,16 +70,72 @@ The graph automatically detects SQL queries in your code:
 - Column-level dead code detection is best-effort.
 - ORM-specific patterns (Prisma, Sequelize, Knex query builder) are not yet supported.
 
+## 🧠 AI Context Layer
+| Tool | Purpose |
+|------|---------|
+| `get_ai_context` | **Boot**: skeleton + docs + compressed files in one call |
+| `get_compressed_file` | Terser-minified source with export legend |
+| `get_project_docs` | Doc Dialect documentation (auto + manual .context/) |
+| `generate_context_docs` | Generate .context/ templates from AST (batch concurrent) |
+| `check_stale_docs` | Detect outdated .ctx files by @sig hash |
+| `discover_sub_projects` | Find sub-projects in monorepo (packages/apps/services/...) |
+| `get_analysis_summary` | Quick health score — cached metrics only, skips cross-file |
+| `compact_project` | Compact all JS files — strips comments/whitespace (mangle:false) |
+| `beautify_project` | Beautify/expand all JS files — inverse of compact |
+
+### AI-First Workflow
+1. **Boot**: `get_ai_context(path)` — loads skeleton + docs (~1700 tokens vs ~60K original)
+2. **Drill**: `expand(symbol)` or `get_compressed_file(file)` — go deeper when needed
+3. **Enrich**: `generate_context_docs(path)` creates `.context/*.ctx` with `{DESCRIBE}` markers. Fill them with compact descriptions.
+
+### Doc Dialect Storage (Two-Tier)
+`.context/` mirrors your source tree with two files per source:
+```
+.context/                  ← auto-generated (mirror)
+├── project.ctx
+├── parser.ctx             ← Machine zone: AST signatures, @sig
+├── parser.ctx.md          ← Agent zone: notes, TODO, decisions
+└── utils/
+    └── helpers.ctx
+
+src/
+├── parser.js
+├── parser.ctx             ← colocated override (wins!)
+└── utils/
+    └── helpers.js
+```
+
+### Doc Dialect Format
+`.context/` files use a compact pipe-separated format:
+```
+--- parser.js ---
+export parseProject()→resolve,findJSFiles,readFileSync|scans dir, parses all files
+parseFileByExtension()→parseSQL,parsePython,parseGo|routes by extension
+PATTERNS: lang-*.js for non-JS|regex fallback for Python
+EDGE_CASES: Python uses regex, not AST|Go interfaces ≠ classes
+```
+
+### Enrichment Workflow
+| Step | How |
+|------|-----|
+| 1. Generate | `generate_context_docs` creates templates with `{DESCRIBE}` markers + `@sig` hash |
+| 2. Enrich | Delegate to agent-pool: `delegate_task({ skill: "doc-enricher" })` |
+| 3. Monitor | `check_stale_docs` detects when source changes invalidate docs |
+| 4. Update | Regenerate with `overwrite: true` — existing descriptions are preserved |
+
 ## 🔍 Code Quality Analysis
 | Tool | Purpose |
 |------|---------|
 | `get_full_analysis` | Run ALL checks + Health Score (0-100) |
+| `get_analysis_summary` | Quick health score (cached only, fast) |
 | `get_dead_code` | Find unused functions/classes |
 | `get_undocumented` | Find missing JSDoc |
 | `get_similar_functions` | Detect code duplicates |
 | `get_complexity` | Cyclomatic complexity metrics |
 | `get_large_files` | Files needing split |
 | `get_outdated_patterns` | Legacy patterns + redundant npm deps |
+| `check_jsdoc_consistency` | Validate JSDoc ↔ AST signatures |
+| `check_types` | Optional tsc type checking (requires TypeScript) |
 | `generate_jsdoc` | Auto-generate JSDoc templates |
 
 ## 🔧 Custom Rules (Configurable)
@@ -96,11 +151,11 @@ Rules are applied automatically based on:
 - Import patterns in source code
 - Code patterns (e.g., `extends Symbiote`)
 
-### Pre-built Rulesets (85 rules)
+### Pre-built Rulesets (86 rules)
 | Ruleset | Rules | Framework |
 |---------|-------|-----------|
 | `symbiote-2x` | 12 | Symbiote.js 2.x |
-| `symbiote-3x` | 17 | Symbiote.js 3.x |
+| `symbiote-3x` | 18 | Symbiote.js 3.x |
 | `react-18` | 6 | React 18 |
 | `react-19` | 5 | React 19 (Server Components) |
 | `vue-3` | 5 | Vue 3 Composition API |
@@ -109,7 +164,7 @@ Rules are applied automatically based on:
 | `fastify-5` | 5 | Fastify 5.x |
 | `nestjs-10` | 6 | NestJS 10.x |
 | `typescript-5` | 5 | TypeScript 5.x |
-| `node-22` | 7 | Node.js 22+ |
+| `node-22` | 13 | Node.js 22+ |
 
 ### Creating New Rules
 Read project workflow docs (e.g., `.agent/workflows/symbiote-audit.md`) and use `set_custom_rule`:
@@ -134,12 +189,14 @@ Read project workflow docs (e.g., `.agent/workflows/symbiote-audit.md`) and use
 |------|---------|
 | `get_filters` / `set_filters` | Configure excluded directories |
 | `add_excludes` / `remove_excludes` | Modify exclude list |
+| `reset_filters` | Reset to defaults |
 
 ## 🚀 Recommended Workflow
 
-1. **Start**: `get_structure` → understand project layout
-2. **Dive**: `get_skeleton` → map code architecture
+1. **Boot**: `get_ai_context` → understand entire project in ~1700 tokens
+2. **Dive**: `expand` / `get_compressed_file` → drill into specific files
 3. **Analyze**: `get_full_analysis` → find issues (Health Score)
 4. **Check Rules**: `check_custom_rules` → framework-specific violations
 5. **Fix**: Address issues by severity (error → warning → info)
 6. **Verify**: `get_pending_tests` → execute in browser → `mark_test_passed/failed` → `get_test_summary`
+7. **Document**: `generate_context_docs` → enrich .ctx files with PATTERNS and EDGE_CASES

package/{AGENT_ROLE_MINIMAL.md → docs/examples/AGENT_ROLE_MINIMAL.md}

@@ -1,6 +1,6 @@
 # Project Graph MCP
 
-You have **Project Graph MCP** tools available — code analysis, project navigation, and framework-specific linting.
+You have **Project Graph MCP** tools available — 45 tools for code analysis, project navigation, monorepo scanning, and framework-specific linting.
 
 ## Quick Start
 
@@ -15,9 +15,11 @@ You have **Project Graph MCP** tools available — code analysis, project naviga
 - `expand(symbol)` — Deep dive into class/function (use symbols from skeleton's `L` field)
 - `deps(symbol)` — Dependency tree (imports, usedBy, calls)
 - `usages(symbol)` — Find all usages across project
+- `invalidate_cache` — Refresh graph after code changes
 
 ### Code Quality
 - `get_full_analysis` — Run ALL checks + Health Score
+- `get_analysis_summary` — Quick health score (cached only, fast)
 - `get_dead_code` — Unused functions/classes  
 - `get_undocumented` — Missing JSDoc
 - `get_similar_functions` — Code duplicates
@@ -25,10 +27,21 @@ You have **Project Graph MCP** tools available — code analysis, project naviga
 - `get_large_files` — Files needing split
 - `get_outdated_patterns` — Legacy patterns
 
+### AI Context
+- `get_ai_context` — **Boot**: skeleton + docs + compressed files (~97% savings)
+- `get_compressed_file` — Terser-minified source with export legend
+- `get_project_docs` — Doc Dialect documentation (.context/)
+- `generate_context_docs` — Generate .context/ templates with `@sig` staleness hashes
+- `check_stale_docs` — Check which .ctx files need updating
+- `discover_sub_projects` — Find sub-projects in monorepo
+- `compact_project` — Compact all JS files (strips comments/whitespace, preserves names)
+- `beautify_project` — Beautify/expand all JS files (inverse of compact)
+
 ### Testing
-- `get_pending_tests` — List @test/@expect annotations
-- `mark_test_passed(testId)` / `mark_test_failed(testId, reason)`
+- `get_pending_tests` — List `[ ]` checklists from `.ctx.md` files
+- `mark_test_passed(testId)` / `mark_test_failed(testId, reason)` — writes to `.ctx.md`
 - `get_test_summary` — Progress report
+- `reset_test_state` — Reset all checklists to `[ ]`
 
 ### Custom Rules
 - `check_custom_rules(path)` — Run framework-specific analysis (auto-detected)
@@ -40,11 +53,13 @@ You have **Project Graph MCP** tools available — code analysis, project naviga
 ## Workflow
 
 ```
-1. get_skeleton("src/")        → Understand structure
-2. get_full_analysis("src/")   → Find issues (Health Score)
-3. check_custom_rules("src/")  → Framework violations
-4. Fix by severity: error → warning → info
-5. get_pending_tests("src/")   → Verification checklist
+1. get_ai_context("src/")       → Boot: ~1700 tokens for entire project
+2. expand(symbol)               → Drill into specific class/function
+3. get_full_analysis("src/")    → Find issues (Health Score)
+4. check_custom_rules("src/")   → Framework violations
+5. Fix by severity: error → warning → info
+6. get_pending_tests("src/")    → Verification checklist
+7. generate_context_docs("src/")→ Enrich .ctx files
 ```
 
 ## Tips

package/package.json

@@ -1,25 +1,24 @@
 {
   "name": "project-graph-mcp",
-  "version": "1.3.0",
+  "version": "2.0.0",
   "type": "module",
-  "description": "MCP server for AI agents — multi-language project graph (JS, TS, Python, Go), code quality analysis, and framework-specific lint rules. Zero dependencies.",
-  "main": "src/server.js",
+  "description": "MCP server for AI agents — project graph, code quality analysis, visual web explorer. JS, TS, Python, Go.",
+  "main": "src/network/server.js",
   "bin": {
-    "project-graph-mcp": "src/server.js"
+    "project-graph-mcp": "src/network/server.js"
   },
   "files": [
     "src/",
+    "web/",
     "vendor/",
     "rules/",
-    "references/",
-    "AGENT_ROLE.md",
-    "AGENT_ROLE_MINIMAL.md",
+    "docs/",
     "CONFIGURATION.md",
     "README.md",
     "LICENSE"
   ],
   "scripts": {
-    "start": "node src/server.js",
+    "start": "node src/network/server.js",
     "test": "node --test tests/*.test.js"
   },
   "keywords": [
@@ -45,6 +44,11 @@
   "author": "RND-PRO",
   "homepage": "https://github.com/rnd-pro/project-graph-mcp",
   "license": "MIT",
+  "dependencies": {
+    "@symbiotejs/symbiote": "^3.2.1",
+    "symbiote-node": "^0.2.0",
+    "ws": "^8.20.0"
+  },
   "engines": {
     "node": ">=18.0.0"
   }