@engrym/mcp-server

The Model Context Protocol server for Engrym — gives your AI coding agent live, structured access to your project's knowledge.

Runs locally on your machine and connects to the Engrym API, exposing your project's decisions, conventions, constraints, and context to any MCP-compatible tool (Claude Code, Cursor, Codex, Gemini CLI). Your agent can query what's been decided and record new knowledge as it works — and because the server runs on your machine and is fully open source, you can audit exactly what it does.

Setup

The simplest path is through the CLI, which writes the right config for your tool:

npx @engrym/cli setup claude-code

The server ships as the engrym-mcp binary and reads your project credential from ~/.engrym/config.yaml. Your tool spawns it as a local subprocess over stdio. setup writes the correct MCP config in each tool's own format and location, so you normally don't edit config by hand — see https://engrym.com/docs/mcp-setup for the per-tool manual configuration.

What your agent can do

The server registers 10 tools — five that read your project's knowledge and five that record it:

• Read — get_team_context (the unified snapshot of decisions, conventions, constraints, and context), query_brain (search the knowledge base), check_conflicts, get_active_intents, get_session_state.
• Write — add_brain_entry (record a new atom), log_decision, broadcast_intent, release_intent, save_session_state.

Each tool's inputs and outputs are typed (Zod schemas), and the tool descriptions are written as agent-facing instructions. The server reaches the Brain only over HTTPS through the Engrym API under your project credential — it runs empty-env and holds no database access of its own.

Docs

https://engrym.com/docs/mcp-setup

License

Apache-2.0 © RTK AI Labs Ltd