Show HN: AutoDocs – Reduce AI costs and never manage context again

2 Aperswal 1 9/14/2025, 6:21:19 PM github.com ↗
Hey all, Adi and Sohan here. We're building Sita (https://trysita.com/), an open-source documentation generator that builds a codebase-aware knowledge graph, writes documentation for all your code, and provides an MCP server so your coding agents get precise, grounded context with citations.

Run 'git clone https://github.com/TrySita/AutoDocs’ and follow our 5 minute set up in the README, and you'll have a local instance of Sita up and running. From there add your repos, use the UI to explore the dependency graph and docs, or plug into an MCP-compatible editor (Codex, Claude Code, Cursor, etc) to fetch context while you code.

Check out a full demo here: https://youtu.be/S-0-9_F1AlE

Before Sita, we felt stuck copy-pasting across files, Slack, and tickets, waiting several minutes while agents guessed about parts of our codebase they were unaware of. With Sita, we generate accurate docs in minutes (which always stay in sync with our code) and feed agents the exact files and functions to change.

There are many code search and mapping tools (e.g., Sourcegraph, CodeSee), but none keep an agent-ready experience with human-readable docs and citations. On the other hand, tools like Notion or Confluence replicate the “docs” experience but lack auto-refresh from code, MCP endpoints, and easy setup.

We wanted our tool to mirror how senior devs navigate real code: by building a mental map of the dependency graph and navigating straight to files and functions. Existing SDKs and hooks don’t support this end-to-end, so we built an open-source pipeline that parses your repo (AST + SCIP), constructs the dependency graph, and generates easy-to-read docs; we expose everything through an MCP server and stream results to the web UI in real time. Our entire stack is open source and free to self host under the Apache 2.0 license at https://github.com/TrySita/AutoDocs. Looking forward to your feedback and hearing your thoughts and comments!

Comments (1)

ssarabu · 2h ago
Cool project, I’ll give it a shot. Curious on what search methods you’re using for the MCP