code-graph-mcp

A high-performance code knowledge graph server implementing the Model Context Protocol (MCP). Indexes codebases into a structured AST knowledge graph with semantic search, call graph traversal, and HTTP route tracing — designed to give AI coding assistants deep, structured understanding of your code.

Features

• Multi-language parsing — Tree-sitter AST extraction across tiers of depth:

  • Full (calls + imports + inheritance + test markers): TypeScript/TSX, JavaScript, Go, Python, Rust, Java. HTTP route extraction additionally covers TypeScript/TSX + JavaScript (Express/Connect), Go (net/http), and Python (Flask/FastAPI) only — Rust and Java web frameworks are not yet route-extracted

  • Smoke-tested (calls + imports + inheritance): C#, Kotlin, Ruby, PHP, Swift, Dart

  • Limited (functions + calls + #include imports + gtest test markers + C++ base-class inheritance; Class::method scope qualification deferred): C, C++

  • Scripting: Bash (functions + commands + source/. imports), Markdown (headings)

  • File-FTS only (no AST symbol extraction): HTML, CSS, JSON

• Semantic code search — Hybrid BM25 full-text + vector semantic search with Reciprocal Rank Fusion (RRF), powered by sqlite-vec

• Call graph traversal — Recursive CTE queries to trace callers/callees with cycle detection

• HTTP route tracing — Map route paths to backend handler functions (Express, Flask/FastAPI, Go net/http)

• Dead code detection — Find unreferenced symbols with smart Orphan/Exported-Unused classification

• Impact analysis — Determine the blast radius of code changes by tracing all dependents

• Incremental indexing — Merkle tree change detection with file system watcher for real-time updates. Smart event filtering skips metadata-only changes (chmod, xattr)

• Context compression — Token-aware snippet extraction for LLM context windows (L0→full code, L1→summaries, L2→file groups, L3→directory overview). Compact JSON output saves 15-20% tokens

• Embedding model — Optional local embedding via Candle (feature-gated embed-model). Context reordered to prioritize structural relations over code for better embedding quality

• Self-healing — Automatic SQLite corruption recovery with rebuild. Startup repair for incomplete indexing (Phase 3 failures)

• MCP protocol — JSON-RPC 2.0 over stdio, plug-and-play with Claude Code, Cursor, Windsurf, and other MCP clients

• Claude Code Plugin — First-class plugin with slash commands (/understand, /trace, /impact), agents, skills, auto-indexing hooks, StatusLine integration, and self-updating

Related MCP server: codebase-memory-mcp

Why code-graph-mcp?

Unlike naive full-text search or simple AST dumps, code-graph-mcp builds a structured knowledge graph that understands the relationships between symbols across your entire codebase.

Incremental by Design

BLAKE3 Merkle tree tracks every file's content hash. On re-index, only changed files are re-parsed — unchanged directory subtrees are skipped entirely via mtime cache. When a function signature changes, dirty propagation automatically regenerates context for all downstream callers across files.

Hybrid Search, Not Just Grep

Combines BM25 full-text ranking (FTS5) with vector semantic similarity (sqlite-vec) via Reciprocal Rank Fusion (RRF) with raw score blending — so searching "handle user login" finds the right function even if it's named authenticate_session. Results are auto-compressed to fit LLM context windows.

Scope-Aware Relation Extraction

The parser doesn't just find function calls — it tracks them within their proper scope context. Extracts calls, imports, inheritance, interface implementations, exports, and HTTP route bindings. Same-file targets are preferred over cross-file matches to minimize false-positive edges.

HTTP Request Flow Tracing

Unique to code-graph-mcp: trace from GET /api/users → route handler → service layer → database call in a single query. Supports Express, Flask/FastAPI, and Go HTTP frameworks.

Zero External Dependencies at Runtime

Single binary, embedded SQLite, bundled sqlite-vec extension, optional local embedding model via Candle — no database server, no cloud API, no Docker required. Runs entirely on your machine.

Built for AI Assistants

Every design decision — from token-aware compression to node_id-based snippet expansion — is optimized for LLM context windows. Works out of the box with Claude Code, Cursor, Windsurf, and any MCP-compatible client.

Performance

Run code-graph-mcp benchmark on your own project to measure.

Efficiency: code-graph vs Traditional Tools

Real-world benchmarks comparing code-graph-mcp tools against traditional approaches (Grep + Read + Glob) on a 33-file Rust project (~537 AST nodes).

Tool Call Reduction

Overall Session Efficiency

What code-graph Uniquely Enables

• Impact analysis — "Changing conn affects 33 functions across 4 files, 78 tests at HIGH risk" — impossible to derive manually with Grep

• Transitive call tracing — Follow main → run_serve → handle_message → handle_tools_call → conn in one query

• Semantic search — Find authenticate_session when searching "handle user login"

• Dependency strength — Not just "file A imports file B", but "file A uses 38 symbols from file B"

When Traditional Tools Are Still Better

Architecture

src/
├── domain.rs     # Shared constants, relation types, env-var config
├── mcp/          # MCP protocol layer (JSON-RPC, tool registry, server)
│   └── server/   # McpServer with IndexingState + CacheState sub-structs
├── parser/       # Tree-sitter parsing, relation extraction, LanguageConfig dispatch
├── indexer/      # 3-phase pipeline, Merkle tree, file watcher
├── storage/      # SQLite schema (v6), CRUD, FTS5, migrations
├── graph/        # Recursive CTE call graph queries
├── search/       # RRF fusion search combining BM25 + vector
├── embedding/    # Candle embedding model (optional, masked mean pooling)
├── sandbox/      # Context compressor with token estimation
└── utils/        # Language detection, config

Installation

Option 1: Claude Code Plugin (Recommended)

Install as a Claude Code plugin for the best experience — includes slash commands, agents, skills, auto-indexing hooks, StatusLine health display, and automatic updates:

# Step 1: Add the marketplace
/plugin marketplace add sdsrss/code-graph-mcp

# Step 2: Install the plugin
/plugin install code-graph-mcp

What you get:

• MCP Server — All code-graph tools available to Claude

• Slash Commands — /understand <module>, /trace <route>, /impact <symbol>

• Code Explorer Agent — Deep code understanding expert via code-explorer

• Auto-indexing Hook — Incremental index on every file edit (PostToolUse)

• StatusLine — Real-time health display (nodes, files, watch status) — compatible with other plugins' StatusLine via composite multiplexer

• Auto-update — Checks for new versions every 6h, updates silently

Manual Update

npm update -g @sdsrs/code-graph

Then reconnect the MCP server in Claude Code with /mcp.

Note: Auto-update is disabled in the source repo directory (dev mode). Use manual update when developing the plugin itself.

Turning auto-update off

Set it in the env block of ~/.claude/settings.json (the environment of the process hosting the MCP server), then reconnect with /mcp.

An update that keeps failing also stops hammering: after 5 failed attempts at the same release the updater drops to one retry per day (and retries immediately when a newer release is published), instead of re-downloading on every session. While it is in that state the statusline shows ⚠ update stuck and code-graph-mcp doctor prints the manual update command.

Invited-memory mode (quieter prompts)

By default, every user prompt the plugin deems code-related gets a small context injection from code-graph CLI output. If you'd rather rely on MEMORY.md + explicit tool calls, opt into invited-memory mode:

1. Adopt the plugin contract into your project (idempotent, self-heals):

   code-graph-mcp adopt

   This writes a sentinel-wrapped managed block into <cwd>/CLAUDE.md (auto-loaded each session) plus <cwd>/.claude/plugin_code_graph_mcp.md (the full decision table, opened on demand). Run code-graph-mcp unadopt to remove both; content outside the managed block is kept.

2. Set the activation env var in ~/.claude/settings.json:

   {
     "env": { "CODE_GRAPH_QUIET_HOOKS": "1" }
   }

3. Restart Claude Code. Session startup skips the project-map injection, UserPromptSubmit stops auto-injecting context, and the MCP instructions become a short pointer to the MEMORY.md file.

Option 2: Claude Code MCP Server Only

Register as an MCP server without the plugin features:

claude mcp add code-graph-mcp -- npx -y @sdsrs/code-graph

Option 3: Cursor / Windsurf / Other MCP Clients

Add to your MCP settings file (e.g. ~/.cursor/mcp.json):

{
  "mcpServers": {
    "code-graph": {
      "command": "npx",
      "args": ["-y", "@sdsrs/code-graph"]
    }
  }
}

Option 4: npx (No Install)

Run directly without installing:

npx -y @sdsrs/code-graph

Option 5: npm (Global Install)

Install globally, then run anywhere:

npm install -g @sdsrs/code-graph
code-graph-mcp

Uninstallation

Claude Code Plugin

# Uninstall the plugin
/plugin uninstall code-graph-mcp

# (Optional) Remove the marketplace
/plugin marketplace remove code-graph-mcp

# (Optional) Clean up all config and cache data
node ~/.claude/plugins/cache/code-graph-mcp/code-graph-mcp/*/scripts/lifecycle.js uninstall

Claude Code MCP Server

claude mcp remove code-graph-mcp

Cursor / Windsurf / Other MCP Clients

Remove the code-graph entry from your MCP settings file (e.g. ~/.cursor/mcp.json).

npm (Global)

Run the teardown first, while the CLI still exists — npm 7 removed the preuninstall/postuninstall lifecycle scripts, so nothing runs on your behalf during npm uninstall. Doing it in the other order leaves the hook entries in ~/.claude/settings.json and the ~40 MB binary cache behind, with no command left on disk to remove them.

code-graph-mcp uninstall     # restore statusline, strip hooks, drop the cache
npm uninstall -g @sdsrs/code-graph

MCP Tools

CLI Commands

All tools are also available as CLI subcommands for shell scripts, hooks, and terminal workflows:

Common options: --json (JSON output), --compact (compact output), --limit N, --depth N, --file <path>.

As of v0.37.0 the CLI is clap-based: every subcommand has --help for its full flag list (code-graph-mcp <command> --help), value flags accept both --flag value and --flag=value, and unknown flags or malformed arguments fail fast with a clear error and a non-zero exit code (2) instead of being silently ignored. For example, trace hides downstream middleware with --no-middleware (shown by default), and snapshot is a create/inspect subcommand pair.

Plugin Slash Commands

Available when installed as a Claude Code plugin:

Supported Languages (19)

Known limitations:

• Kotlin/Swift interface conformance is recorded as inherits (both use a single : Type grammar for base classes and protocols/interfaces), so implements-filtered queries return empty for these two languages.

• Cross-file dead-code detection may false-positive a type whose only cross-file reference sits beyond the 4096-byte stored-content cap per node (documented accepted limitation, v0.97.1).

Team-shared graph snapshot

Skip the full local index for team members and CI runners by publishing a ~3-5MB graph snapshot with each GitHub release.

Setup (one-time):

1. Copy node_modules/code-graph-mcp/templates/code-graph-snapshot.yml into your repo's .github/workflows/.

2. Push a release tag. The workflow uploads code-graph-snapshot-<sha>.db.zst as a release asset.

Verify:

npx code-graph-mcp snapshot inspect ./code-graph-snapshot-<sha>.db.zst

After setup, the auto-fetch is opt-in per consumer: an untrusted repo could otherwise seed a misleading graph, so an unconfigured clone prints a hint and skips the install. Enable it with one of the trust signals below. These live in the environment (never in .code-graph.toml) so a committed/PR-injected config file cannot set them.

Integrity fail-closes: with no pin set and no fetchable <url>.blake3 sidecar, install is refused rather than accepting unverified content. Set a pin, or have the publisher serve the .blake3 sidecar alongside the snapshot. (file:// sources, used in tests, are exempt.)

Offline / air-gapped usage

The optional embedding model (needed for vector semantic search) is downloaded lazily on first use. To control that in a restricted environment:

Installing the model by hand

CODE_GRAPH_MODEL_DIR is the supported manual route. Hand-populating the default cache dir (<platform cache>/code-graph/models) does not work: that dir is only trusted once extract_and_promote has written a .model-id marker, so a manually filled copy is treated as not-current and re-downloaded.

# Download + verify against the published checksum, then extract flat
# (the archive has no wrapping folder: model.safetensors, tokenizer.json, config.json).
curl -sL -o models.tar.gz \
  https://github.com/sdsrss/code-graph-mcp/releases/download/v<version>/models.tar.gz
curl -sL https://github.com/sdsrss/code-graph-mcp/releases/download/v<version>/models.tar.gz.sha256
sha256sum -c models.tar.gz.sha256
mkdir -p ~/.cache/code-graph/models-manual
tar -xzf models.tar.gz -C ~/.cache/code-graph/models-manual
export CODE_GRAPH_MODEL_DIR=~/.cache/code-graph/models-manual

On Windows use %LOCALAPPDATA%\code-graph\models-manual, and give tar a POSIX path for -C (/c/Users/<you>/...) — tar -C C:\... fails with "Cannot connect to C: resolve failed". The model is pinned by content hash, so a manual install must be redone on each version bump.

Diagnosing a model that never downloads

code-graph-mcp doctor reports the last download outcome, distinguishing "no download has been attempted" from "download FAILED after N attempt(s): …" with the underlying error. The raw record is <platform cache>/code-graph/model-download.json. On a network that performs TLS inspection, the download retries automatically against the OS certificate store after the bundled roots are rejected.

Storage

Uses SQLite with:

• FTS5 for full-text search

• sqlite-vec extension for vector similarity search

• Merkle tree hashes for incremental change detection

Data is stored in .code-graph/index.db under the project root (auto-created, gitignored).

Build from Source

Prerequisites

• Rust 1.75+ (2021 edition)

• A C compiler (for bundled SQLite / sqlite-vec)

Build

# Default build — FTS5-only (~10 MB binary, no embedding model)
cargo build --release

# With local embedding model (~150 MB binary; downloads ~120 MB model lazily on first semantic search)
cargo build --release --features embed-model

Direct cargo install users get the FTS5-only build by default. npm/npx/plugin users get the full hybrid (FTS5 + vector) build automatically — release CI compiles with --features embed-model for shipped binaries.

Configure (from source)

Add the compiled binary to your MCP settings:

{
  "mcpServers": {
    "code-graph": {
      "command": "/path/to/target/release/code-graph-mcp"
    }
  }
}

Development

# Run tests
cargo test

# Run tests without embedding model
cargo test --no-default-features

# Check compilation
cargo check

# Run performance benchmarks (indexing, search, call graph)
cargo bench --no-default-features

License

See LICENSE for details.