From repo to live graph
Four steps to give your agent full-stack context — the code, and what it actually does when it runs. Neat is free to self-host, Apache 2.0.
On Windows, install once with npm i -g neat.is and run neat — npm
writes the shim as a literal neat.is file, which Windows won't execute.
Run Neat in your repo
npx neat.is Neat reads your codebase with tree-sitter and builds the graph — services, files, calls, databases, infrastructure — then starts a local daemon that keeps it current as you commit. No config, no manual diagramming.
Point your agent at the graph
The graph is queryable over MCP (Model Context Protocol) and a REST API. Connect your coding agent's MCP client to the daemon and it reads the whole map instead of grepping around.
Add the runtime layer
The OBSERVED edges — what your system actually does — come from your live runtime. Emit OpenTelemetry from your app, or add a connector for a provider you already run. Now the graph shows what runs, not only what the code declares.
Browse connectors →Ask the graph
Query it through the MCP tools: divergence (where declared and observed reality disagree), blast radius (everything a change touches), dependencies, and policy checks. The answer is a traversal, not a guess.
Full reference, configuration, and the source live on GitHub.
Read the docs on GitHub ↗