AutoDocs MCP
An MCP server that graphs repository dependencies, writes docs from the graph and answers agent questions via codebase-qna
- 196 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- Sends repository content to external model APIs for summaries and embeddings and needs API keys
- Starts local services in Docker with a database and open ports 3000 and 8000
Install
- Works in
Manual install
cp .env.example .env
docker compose up -dFill in the summary and embedding keys in .env first. Docker with Compose, pnpm 10 or newer and uv are required. The UI opens at http://localhost:3000.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Sends repository content to external model APIs for summaries and embeddings and needs API keys
- Starts local services in Docker with a database and open ports 3000 and 8000
This is third-party code. Review the repository files before installing.
What it does
AutoDocs parses a repository with tree-sitter and SCIP, builds a graph of files, definitions, calls and imports, sorts it topologically and generates repository-wide documentation by walking that graph. Everything runs locally in Docker Compose: a PostgreSQL database, a FastAPI API (port 8000) and a Next.js web UI (port 3000) with chat and code browsing. The MCP server is built into the web app and listens at http://localhost:3000/api/mcp. It exposes the codebase-qna tool, which answers questions about a chosen repository by querying the analysis databases. The repository is selected with the x-repo-id header, and the ID comes from the UI. Summaries and embeddings are produced through OpenAI-compatible APIs: OpenRouter and OpenAI by default. Setup guides for Claude, Cursor and Continue live at docs.trysita.com. The hosted and enterprise versions of Sita are on a waitlist for now.
Who it is for. Developers who run AutoDocs themselves and want Claude Code, Cursor or another MCP client to find answers about a large repository's structure without rereading files.
Good fit when
- An agent struggles in a large repository and burns context reading files by hand
- You want documentation built from the real links between files, calls and imports
- The repository is TypeScript, JavaScript or Python and you want a local UI with chat over the code
- An MCP client needs access to several repositories, chosen through the x-repo-id header
Not a fit when
- The repository is Go, Kotlin, Java or Rust: support for these languages is planned but absent today
- One repository mixes several languages: polyglot projects are not supported yet
- The code sits in a nested folder of the repository: the tool expects it at the root
- You want a ready cloud service without Docker: the hosted Sita version is waitlist-only
Example request
Use codebase-qna to find which modules call the payment function and show the call chain up to the request handlerMCP
- Transport
- http
- Authentication
- not required
Environment variables
SUMMARIES_API_KEYrequired, secret- Key of the OpenAI-compatible API used to generate summaries.
SUMMARIES_MODELrequired- Model used for summaries.
SUMMARIES_BASE_URL- Base URL of the summaries API, OpenRouter by default.
EMBEDDINGS_API_KEYrequired, secret- Key of the OpenAI-compatible API used for embeddings.
EMBEDDINGS_MODELrequired- Model used for embeddings.
EMBEDDINGS_BASE_URL- Base URL of the embeddings API, OpenAI by default.
GITHUB_TOKENsecret- GitHub token for private repositories and API rate limits.
MAX_REQUESTS_PER_SECOND- Request cap for batched summary generation, 15 by default.
Limitations
- Running it needs Docker with Compose, pnpm 10 or newer and uv, plus keys for OpenAI-compatible APIs for summaries and embeddings: external models process the repository content.
- TypeScript, JavaScript and Python are supported, polyglot repositories are not, and the code must sit at the repository root.
- Updating is manual: you delete the repository in the workspace and add it again, with a resync button planned.
- The latest commit is from June 2026.
How to disable. Remove the AutoDocs entry from the MCP client settings and stop the services with `docker compose down` in the repository directory. The local analysis databases go away with the Docker volumes.
FAQ
How do I connect AutoDocs to Claude Code or Cursor?
Start the services with `docker compose up -d`, add a repository in the UI at localhost:3000 and point the client at http://localhost:3000/api/mcp with the x-repo-id header. The repository ID is shown in the UI, and step-by-step guides for Claude, Cursor and Continue live at docs.trysita.com.
What tool does the AutoDocs MCP server provide?
One: codebase-qna. It answers questions about the chosen repository using the analysis databases AutoDocs built at ingestion.
How do I refresh the docs after code changes?
Manually for now: delete the repository in the workspace and add it again, and ingestion starts automatically. A resync button and periodic ingestion are planned.
Which languages does AutoDocs support?
TypeScript, JavaScript and Python. Go, Kotlin, Java and Rust are planned, and polyglot repositories are not supported yet.
Which keys does AutoDocs need?
A key and model for summaries and a key and model for embeddings, with OpenRouter and OpenAI as defaults. A GitHub token is needed only for private repositories or when API rate limits bite.
Discussion
No questions or reviews yet.
Related
MCP serversDocs MCP Server
Grounded Docs MCP Server
A local MCP server that indexes docs from websites, GitHub, npm, PyPI and files and gives the agent version-aware search
An MCP server that turns a large codebase into a semantic feature graph for precise navigation and meaning-based search
Ref's MCP server: documentation search for APIs and libraries that hands the agent only the relevant parts of pages
Deep Graph MCP: a code graph of a repository for agents
Deep Graph MCP
An MCP server that gives an agent a repository code graph: search by meaning, symbol links and the places an edit would touch
A local code graph with version history: an MCP server that tells the agent who calls a function and what an edit breaks
An MCP server that answers an agent's questions from the documentation of one Rust crate at an exact version
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| AutoDocs (this card) | yes | not required | 196 | |
| codebase-memory-mcp | no | not required | 48,228 | builds a code graph locally for structural queries, while AutoDocs also generates documentation with a chat in a web UI |
| Deepwiki MCP | no | not required | 1,389 | fetches ready pages from deepwiki.com, while AutoDocs parses your own repository on your machine |
README in short
The README explains what the repository does, lists the prerequisites, environment variables and the Docker Compose launch, then shows the MCP address and header. It closes with the project layout, troubleshooting and known limitations.