AutoDocs

Follow Us

• Sohan - @SohanSarabu
• Adi - @adiperswal

Docs · Report Bug · Feature Request ·

Sita uses GitHub Discussions for Support and Feature Requests.

Your browser does not support the video tag.

AutoDocs, by Sita

Automate documentation for any repo: we traverse your codebase, parse the AST, build a dependency graph, and walk that graph to generate accurate, high-signal docs. A built-in MCP server lets coding agents deep-search your code via HTTP.

(Interested in our hosted or enterprise offerings? Join the waitlist at https://trysita.com)

What This Repo Does

• Parses your repository using tree-sitter (AST parsing) and SCIP (symbol resolution).
• Constructs a code dependency graph (files, definitions, calls, imports) and topologically sorts it.
• Traverses that graph to create repository-wide, dependency-aware documentation and summaries.
• Exposes a FastAPI backend for ingestion/search and a Next.js web UI for chat and exploration.
• Provides an MCP server so agentic tools can query your repo with deep search.

---

Prerequisites

Install these once on your machine:

• pnpm 10+ (Node 20+ recommended; Corepack is fine). Docs
• uv (fast Python package manager). Docs
• Docker + Docker Compose (to run everything locally). Docs

Reference docs

• pnpm install: https://pnpm.io/installation
• uv install: https://docs.astral.sh/uv/

GitHub Personal Access Token (optional)

Some features or scripts may call the GitHub API (e.g., fetching repo metadata). If you hit rate limits or need to access private repos, create a Personal Access Token (PAT) and set it in your environment.

• How-to (official docs): https://docs.github.com/en/authentication/keeping-your-account-and-data-secure/creating-a-personal-access-token
• Create fine-grained PAT (recommended): https://github.com/settings/personal-access-tokens/new
• Create classic PAT (legacy): https://github.com/settings/tokens/new

Suggested scopes

• Public repos only: use a fine-grained token with selected repositories (read-only) or a classic token with public_repo.
• Private repos: fine-grained token with read-only repo access to the needed repositories, or a classic token with repo.

Add to your .env (or shell env):

GITHUB_TOKEN=ghp_your_token_here

Notes

• Keep this token secret; do not commit .env.
• Fine-grained tokens are preferred for tighter, per-repo permissions.

Quickstart (copy/paste)

1. Environment

cp .env.example .env

Configuration

Preconfigured

• Database: DATABASE_URL (local postgres DB). In Compose, DB is available at postgresql://postgres:postgres@db:5432/app.
• Ingestion API: INGESTION_API_URL for the web app to call the FastAPI service.
• Analysis storage: ANALYSIS_DB_DIR controls where generated per-repo SQLite files live.

---

To-be configured

• Summaries: SUMMARIES_API_KEY, SUMMARIES_MODEL, SUMMARIES_BASE_URL (OpenAI-compatible, default OpenRouter)
• Embeddings: EMBEDDINGS_API_KEY, EMBEDDINGS_MODEL, EMBEDDINGS_BASE_URL (OpenAI-compatible, default OpenAI)
• Rate limiting: MAX_REQUESTS_PER_SECOND for LLM summary batching (default 15)

---

1. Run locally with Docker Compose

docker compose up -d

# If you want to see logs
docker compose up

You should now have:

• Web UI at http://localhost:3000
• API at http://localhost:8000 (OpenAPI schema at /schema)

Updating Docs

To refresh a repository’s docs after code changes, remove the repo and re-ingest it (temporary workflow):

• UI: delete the repo in your Workspace, then add it again (ingestion starts automatically).

We’re actively adding a one-click "Resync" button in the UI, followed by automatic periodic ingestion (coming soon)

Using the MCP Server

The MCP server is available at http://localhost:3000/api/mcp and is designed for coding agents and MCP-compatible clients. It exposes a codebase-qna tool that answers repository-scoped questions by querying the analysis databases that AutoDocs produces.

Tips

• Point your MCP client at http://localhost:3000/api/mcp.
• Include an x-repo-id header with the repo ID (you can find it in the UI).
• For setup guides with popular tools (Claude, Cursor, Continue), see https://docs.trysita.com

Development Workflow (for contributing)

For a local dev loop without Docker Compose you can run the API and web dev servers directly:

# concurrent dev (API + Web + DB)
./tools/dev.sh --sync

Database migration (run if modifying the postgres schema)

cd packages/shared
DATABASE_URL=postgresql://postgres:postgres@localhost:5432/app pnpm drizzle-kit push --config drizzle.main.config.ts

Project Layout

• ingestion/ — Python FastAPI service, AST parser, graph builder, embeddings, and search.
• webview/ — Next.js app (Turborepo workspace) and shared TS packages.
• docker-compose.yml — local Postgres, API, and Web services.
• tools/ — helper scripts (build_all.sh, dev.sh, uv_export_requirements.sh).

Troubleshooting

• Web can’t reach the API: ensure INGESTION_API_URL=http://api:8000 is set in .env.
• Missing uv/pnpm: install them (see links above)

Known Issues

• In your repositories, code must live at the repository root, not in a nested folder.
• Language support: currently supports TS, JS, and Python; currently working on expansion to Go, Kotlin, Java, and Rust.
• Polyglot repos (multiple languages in one repo): not supported yet, but we’re actively working on it.

License

Licensed under the Apache 2.0 License. See [LICENSE](./LICENSE).

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: TrySita
• Source: TrySita/AutoDocs
• License: Apache-2.0
• Homepage: https://trysita.com/

Install and usage instructions live in the source repository linked above.