Skip to main content
MCP server CX Codex

engram · nickcirv — Claude Code MCP server: code graph and project context

Builds a local graph of your codebase and hands the agent the relevant files and links instead of everything at once. Fits projects that no longer fit in context.

Install into Claude Code · Platform CLI All methods and platforms ↓
claude mcp add --env ENGRAM_API_TOKEN=<ENGRAM_API_TOKEN> --env ENGRAM_MISTAKE_GUARD=<ENGRAM_MISTAKE_GUARD> --env ENGRAM_ANTHROPIC_MEMORY_PATH=<ENGRAM_ANTHROPIC_MEMORY_PATH> --env ENGRAM_MCP_CONFIG_PATH=<ENGRAM_MCP_CONFIG_PATH> --env ENGRAM_NO_UPDATE_CHECK=<ENGRAM_NO_UPDATE_CHECK> --transport stdio nickcirv-engram -- npx -y engramx
Open on GitHub

What engram does

Engram indexes source entities, references, Git co-change patterns and recorded mistakes into a project-local graph. The agent queries it and gets back bounded context with an estimated token count, then opens only what matters. The index lives in the project's .engram/ directory and persists through sql.js on SQLite.

You reach it through a CLI, MCP, an authenticated local dashboard and hooks. Hooks intercept file reads and greps, which is why plain init installs one by default. Run it with --no-hook to inspect the graph output before you change how your agent behaves.

Node 20 or later is required, and the package is engramx. The system is not guaranteed to stay off the network: the CLI checks npm for updates, and configured providers or plugins can spawn their own processes. Set ENGRAM_NO_UPDATE_CHECK=1 to stop the update check. The HTTP server binds to 127.0.0.1, validates Host and Origin, and wants a bearer token from ENGRAM_API_TOKEN or a local token file.

What it helps with

  • locate the right code before opening whole files
  • trace who calls a function and what depends on it
  • carry past debugging mistakes into a new session
  • generate IDE rule files from one index
  • estimate how many tokens a context request costs

Example prompts for the agent

  • Query the graph for everything related to authentication and show me only the files I need
  • Index this project without the hook, I want to see the output first
  • Show me the callers of this function before I change it

How to install engram

Platform CLI

claude mcp add --env ENGRAM_API_TOKEN=<ENGRAM_API_TOKEN> --env ENGRAM_MISTAKE_GUARD=<ENGRAM_MISTAKE_GUARD> --env ENGRAM_ANTHROPIC_MEMORY_PATH=<ENGRAM_ANTHROPIC_MEMORY_PATH> --env ENGRAM_MCP_CONFIG_PATH=<ENGRAM_MCP_CONFIG_PATH> --env ENGRAM_NO_UPDATE_CHECK=<ENGRAM_NO_UPDATE_CHECK> --transport stdio nickcirv-engram -- npx -y engramx

Replace <…> with your own values

Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more

Questions about engram

How do I install engram in Claude Code?

Run claude mcp add engram -e ENGRAM_API_TOKEN=<ENGRAM_API_TOKEN> -- npx -y engramx. To pass the full set of variables instead: `claude mcp add --env ENGRAM_API_TOKEN=<ENGRAM_API_TOKEN> --env ENGRAM_MISTAKE_GUARD=<ENGRAM_MISTAKE_GUARD> --env ENGRAM_ANTHROPIC_MEMORY_PATH=<ENGRAM_ANTHROPIC_MEMORY_PATH> --env ENGRAM_MCP_CONFIG_PATH=<ENGRAM_MCP_CONFIG_PATH> --env ENGRAM_NO_UPDATE_CHECK=<ENGRAM_NO_UPDATE_CHECK> --transport stdio nickcirv-engram -- npx -y engramx`.

What does engram do?

It indexes code entities, Git relationships and notes on past mistakes, then serves narrow context to the agent on request. The graph is stored in .engram/ using sql.js over SQLite.

Do I need an API token?

Yes, for the protected routes of the HTTP server and the dashboard. It is read from ENGRAM_API_TOKEN or a local token file.

Similar MCP servers

All MCP servers for Agent setup: memory, context, orchestration →

engram by other authors comes first.

engram tstockham96 Stores project knowledge locally and helps an agent retrieve it by meaning. Useful when context needs to carry across sessions and projects. CX Codex +1 ★ 47repo stars Engram get-engram Keeps every agent conversation word for word and finds it again by meaning. One shared memory across Claude, Codex, Cursor, Copilot and Zed. CX Codex +1 ★ 19repo stars REA morluto Analyzes binaries, Electron bundles, and JavaScript runtimes directly from your agent. It inspects compiled apps without source code to recreate specific features. CX Codex +1 ★ 386,989repo stars GitHub official github Wires your AI agent into GitHub: read code, open issues and pull requests, watch Actions runs. Saves a trip to the browser. CX Codex +1 ★ 33,918repo stars Netdata netdata Hands your AI agent live monitoring data: metrics, logs, alerts and anomaly detection. Useful when you need to know what a server is doing without opening the dashboard. CX Codex +1 ★ 81,115repo stars browser-use browser-use Give an agent browser tasks such as finding information, extracting page data, and filling forms. Connect it when those actions need to happen on the web. CX Codex +1 ★ 119,160repo stars Codebase Memory DeusData A local MCP server that maps your repo's code, docs and infrastructure into a graph. Gives the agent structural context without re-reading files. CX Codex +1 ★ 47,392repo stars mcp-server HeyPuter Connects your AI agent to Puter for files, websites and workers. Useful when the project already lives in Puter's cloud. CX Codex +1 ★ 43,872repo stars

Project sponsors

Companies that keep this analytics open
Gold sponsor

Impressions, clicks and code copies — open statistics. We keep the numbers public for the same reason NeBlask itself is public. Become a sponsor