Install
openclaw skills install codegraph-toolCodeGraph — Pre-indexed code knowledge graph tool. Provides semantic code intelligence for Claude Code, Cursor, Codex, Gemini, OpenCode, Hermes, Antigravity, Kiro and other AI agents.
openclaw skills install codegraph-toolLocal-first code intelligence tool. tree-sitter parsing → SQLite (FTS5) storage → MCP exposes knowledge graph.
Use when users need to install CodeGraph, initialize code graph, index projects, understand code structure, trace call chains, evaluate change impact.
npx @colbymchenry/codegraph # Recommended: one-click install + auto-configure Agent
curl -fsSL https://raw.githubusercontent.com/colbymchenry/codegraph/main/install.sh | sh # No Node required
npm i -g @colbymchenry/codegraph # npm
Non-interactive: codegraph install --yes [--target=claude,cursor] [--location=local]
cd your-project
codegraph init -i # Create .codegraph/ + build index
codegraph index # Full index (or rebuild: --force)
codegraph sync # Incremental sync (changed files only)
codegraph status # Check status
Tool selection guide is automatically sent to Agent in MCP
initializeresponse.
| Intent | Tool | Key Parameters |
|---|---|---|
| Almost any question — architecture, flow, bug, investigation | codegraph_explore ⭐ | query (natural language or symbol names) |
| Locate symbol by name | codegraph_search | query, kind, limit |
| Call chain | codegraph_callers / codegraph_callees | symbol, limit |
| Change impact | codegraph_impact | symbol, depth |
| Single symbol full source / overload disambiguation | codegraph_node | symbol, includeCode, file, line |
| File structure | codegraph_files | — |
| Index health | codegraph_status | — |
codegraph_explore is the PRIMARY tool — call it first, usually the only call needed. Returns relevant symbol source code (grouped by file), with line numbers.explore > search + node chained calls.❌ Don't use grep to verify codegraph results (AST is more accurate than grep)
❌ Don't chain search + node to understand an area (one explore does it)
❌ Don't loop node over multiple symbols (one explore returns all)
⚠️ Check staleness banner after edits — pending files need direct Read
Three-layer protection: ① Native FS events + 2s debounce auto-sync ② Per-file staleness banner (pending files prefixed with ⚠️) ③ Auto-reconciliation on MCP reconnect.
Manual sync only needed in sandbox (CODEGRAPH_NO_DAEMON=1) or CI pre-flight.
codegraph query <symbol> [--kind class] [--limit 10] [--json]
codegraph callers <symbol> [--limit 20] [--json]
codegraph callees <symbol> [--limit 20] [--json]
codegraph impact <symbol> [--depth 2] [--json]
codegraph context "<task>" [--max-nodes 20]
codegraph affected <files...> [--stdin] [--depth 5] [--filter "e2e/*"] [--quiet]
Zero-config — no config files. Auto-excludes node_modules/vendor/dist/build/target/.venv/Pods/.next + .gitignore entries + >1MB files. Exclude more: add to .gitignore; bring back: !vendor/.
Environment variables: CODEGRAPH_WATCH_DEBOUNCE_MS(2000) CODEGRAPH_NO_DAEMON CODEGRAPH_EXPLORE_LINENUMS(1) CODEGRAPH_ADAPTIVE_EXPLORE(1)
TS/JS/Python/Go/Rust/Java/C#/PHP/Ruby/C/C++/Swift/Kotlin/Scala/Dart/Svelte/Vue/Liquid/Pascal-Delphi/Lua/Luau
Django/Flask/FastAPI/Express/NestJS/Laravel/Drupal/Rails/Spring/Gin-chi-gorilla-mux/Axum-actix-Rocket/ASP.NET/Vapor/ReactRouter-SvelteKit — auto-recognize route files and generate route nodes.
Swift↔ObjC / RN legacy bridge / TurboModules / RN events / Expo Modules / Fabric-Paper views. Synthesized edges marked provenance:'heuristic'.
codegraph uninstall # Remove MCP config
codegraph uninit # Delete .codegraph/