CodeGraph
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@CodeGraphfind the function that validates email addresses"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
CodeGraph
CodeGraph turns source code and project knowledge into a searchable graph for AI agents and developers. It parses code with tree-sitter, stores structural and temporal relationships in FalkorDB, and exposes five focused tools through the Model Context Protocol (MCP).
Access the project
The public npm package is @agntk/codegraph-mcp. Version 0.1.0 provides the codegraph-mcp and codegraph-dashboard bins.
Related MCP server: ContextAtlas
What it does
Searches code by symbol, structure, and meaning.
Extracts functions, classes, interfaces, variables, types, imports, calls, and other relationships.
Stores project facts with
valid_atandinvalid_attimestamps for point-in-time queries.Supports first-class TypeScript, JavaScript, Python, Go, Rust, and Markdown parsing, with generic tree-sitter support for additional languages.
Runs with embedded FalkorDBLite on Linux x64 and Apple silicon macOS, or an external FalkorDB service on any supported Node.js platform. Apple silicon requires Homebrew
libompandopenssl@3runtime libraries.Explores exact graph totals in Files or Symbols view with most-connected-first windows, Previous/Next paging, append-style loading, in-place neighbor expansion, and a Files-view toggle for unresolved external modules.
Exposes
analyze,codebase,knowledge,query, andsearchMCP tools.
Choose how to start
CodeGraph requires Node.js 20 or newer. The npm commands below install the public @agntk/codegraph-mcp package from the registry.
Agent-first with MCP
Add this server configuration to an MCP client:
{
"mcpServers": {
"codegraph": {
"command": "npx",
"args": ["-y", "@agntk/codegraph-mcp"]
}
}
}In the first conversation, ask the agent to check CodeGraph status:
{"method":"tools/call","params":{"name":"codebase","arguments":{"action":"status"}}}On a fresh install, codebase status returns configured: false and setupRequired: true. An empty database is a valid first run. Configure one absolute project path, then start a full index:
{"method":"tools/call","params":{"name":"codebase","arguments":{"action":"configure","projectAction":"set","projects":["/absolute/path/to/project"]}}}{"method":"tools/call","params":{"name":"codebase","arguments":{"action":"reindex","mode":"full","scope":"/absolute/path/to/project"}}}Configuration saves the project but does not index it. The full reindex parses structure first and then finishes embeddings. With the local provider, codebase status includes model load and indexing state while work is running.
Dashboard-first in a browser
Start the dashboard directly from the package:
npx -y -p @agntk/codegraph-mcp codegraph-dashboardOpen the URL printed by the process. A fresh database opens on the setup flow. Confirm storage and embeddings, choose Browse to select a folder, then select Index project. The page shows model download and indexing progress, finishes remaining embeddings automatically, and reports file, symbol, edge, and embedding counts before opening the graph explorer.
The dashboard and MCP server are separate entry points. They can run at the same time against the same data path.
Platform and storage
Platform | Default driver | Prerequisites | Configuration |
Linux x64 | Embedded FalkorDBLite | Node.js 20 or newer | No database configuration when the embedded package is available |
macOS Apple silicon | Embedded FalkorDBLite | Node.js 20 or newer, | No database configuration when the libraries and embedded package are available |
macOS Intel | External FalkorDB | Node.js 20 or newer | Set the external variables below |
Windows x64 | External FalkorDB | Node.js 20 or newer | Set the external variables below |
Linux arm64 and other platforms | External FalkorDB | Node.js 20 or newer | Set the external variables below |
An explicit CODEGRAPH_DRIVER always wins. Without it, a configured FALKORDB_URL or FALKORDB_HOST selects external FalkorDB. Otherwise CodeGraph selects embedded FalkorDBLite on Linux x64, or on Apple silicon macOS when its native libraries are present, and falls back to external FalkorDB if the embedded package cannot load. All other platforms default to external FalkorDB.
CODEGRAPH_DATA_DIR defaults to ~/.codegraph; the MCP project configuration is stored at ~/.codegraph/mcp-context.json. Embedded database storage separately defaults to <current working directory>/.codegraph/falkordb. If that database path is too long for a Unix socket, CodeGraph relocates it to ~/.codegraph/graphs/<12-hex-digest>. Set CODEGRAPH_DATA_DIR to a writable directory for MCP configuration, and set CODEGRAPH_DB_PATH to a writable directory for embedded database files.
For external FalkorDB, set either a URL:
CODEGRAPH_DRIVER=falkordb
FALKORDB_URL=host:6379Or set a host and port:
CODEGRAPH_DRIVER=falkordb
FALKORDB_HOST=host
FALKORDB_PORT=6379One database server runs per data directory; a second CodeGraph process attaches to it.
Embeddings and privacy
Provider | Model and dimensions | Configuration | First run |
Local |
| No key and no provider override | Default no-key path. Downloads approximately 132 MiB, measured as 138,011,417 bytes (131.6 MiB), shows progress, then caches the model locally. |
Voyage |
|
| Sends embedding inputs to Voyage. |
OpenRouter |
|
| Sends embedding inputs to OpenRouter. |
None | No model, 0 |
| Explicit structural-only mode. Semantic similarity search is unavailable. |
If no provider is set, a configured Voyage key selects Voyage, a configured OpenRouter key selects OpenRouter, and no key selects local. Set CODEGRAPH_EMBEDDING_PROVIDER explicitly when you need to pin the choice.
Provider, model, and dimension are stored with the graph. If any of them changes, CodeGraph blocks graph mutation with this remedy: Run an explicit re-embed migration or a full reindex before using the requested embedding profile.
MCP tools
Tool | Purpose |
| Inspect impact, call hierarchy, cycles, dead-code candidates, hotspots, change coupling, and ownership inferred from indexed git history |
| Find code and project knowledge by name, structure, or meaning |
| Store and recall entities, relationships, facts, and documents |
| Configure projects, index code, inspect status, and read source |
| Run read-only Cypher queries against the graph |
Set CODEGRAPH_RAW_TOOLS=true to expose lower-level handlers alongside the five grouped tools.
Configuration
Variable | Purpose |
|
|
| MCP configuration directory, default |
| Embedded FalkorDBLite data path, default |
| External FalkorDB host and port |
| External FalkorDB host, default |
| External FalkorDB port, default |
| Graph name, default |
|
|
| Voyage embeddings and optional reranking |
| OpenRouter embeddings or LLM features |
| Optional LLM-backed knowledge extraction |
| Dashboard and REST API port, default |
| Additional absolute dashboard Browse roots, separated by commas |
| Set to the literal value |
Never commit provider keys. Put secrets in the MCP client's protected environment configuration or a local ignored .env file.
Release verification
The canonical package command is:
pnpm pack:npmIt builds and validates one publishable tarball under tmp/release. The complete release gate is:
pnpm release:checkThe basic installed-tarball smoke proves the package installs in a clean consumer, both binaries start, all five MCP tools are present, a fresh database is setup-safe, Browse and indexing work, data survives restarts, the dashboard and MCP process share one embedded server, and the exact tarball hash is reported. The release CI runs that installed artifact with embedded storage on Linux x64 and Apple silicon, and checks external FalkorDB guidance on Windows x64. An opt-in local-provider lane proves cold model download progress and a usable 768-dimension vector index.
A clean-consumer local tarball proof verifies that bare npx -y @agntk/codegraph-mcp selects the codegraph-mcp bin and that npx -y -p @agntk/codegraph-mcp codegraph-dashboard selects the dashboard bin. Both forms also resolve and boot from the public npm registry.
Development
pnpm install --frozen-lockfile
pnpm audit:prod
pnpm lint
pnpm typecheck
pnpm test
pnpm build
pnpm docker:db
pnpm test:integration
pnpm test:scripts
pnpm build:mcpb
pnpm release:checkSee CONTRIBUTING.md for the pull request workflow and required evidence.
FalkorDBLite's Linux x64 and Apple silicon macOS binaries are installed with the workspace dependencies. Apple silicon also requires Homebrew libomp and openssl@3.
Packages
Package | Responsibility |
Indexing, search pipelines, services, and git synchronization | |
FalkorDB and FalkorDBLite drivers, graph queries, and knowledge operations | |
Embeddings, reranking, entity resolution, and document ingestion | |
MCP transport and the five public tool groups | |
Source-checkout command-line tools | |
Public npm distribution staging and entry point | |
Platform-local MCPB desktop extension build | |
REST API consumed by the dashboard | |
Static dashboard UI served by the API |
The marketing site lives in apps/web, and the reproducible search benchmark lives in benchmarks/cgbench-v1.
License
CodeGraph is available under the MIT License.
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Servers
- AlicenseAqualityAmaintenanceProvides AI coding agents with five intelligence layers (dependency graph, git history, documentation, architectural decisions, code health) via nine MCP tools, enabling deep codebase understanding and reducing exploration cost.116,273AGPL 3.0
- AlicenseNot gradedqualityCmaintenanceEnables AI coding agents to retrieve and manage code context with hybrid search, project memory, and observability via MCP tools.29MIT
- AlicenseNot gradedqualityAmaintenanceProvides code intelligence for AI coding agents by indexing repositories into a hybrid knowledge graph, enabling agents to query dependencies, impact, and context through 28 MCP tools.3Apache 2.0
- AlicenseNot gradedqualityDmaintenanceEnables AI coding assistants to semantically search and retrieve relevant code patterns, documentation, and implementations from a codebase via MCP tools.6MIT
Related MCP Connectors
Code intelligence for coding agents: semantic, AST, graph, and full-text search. 279+ languages.
Your memory, everywhere AI goes. Build knowledge once, access it via MCP anywhere.
Codebase intelligence for agents: 152 structured artifacts across 21 programs, one call.
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/Phoenixrr2113/codebase-graph'
If you have feedback or need assistance with the MCP directory API, please join our Discord server