docchat-server 0.0.1

Version-pinned documentation retrieval as a Model Context Protocol server. Gives Claude Code / Cursor / any MCP-aware AI grounded answers from the docs of the exact library version your lockfile pins.

docchat-server

Version-pinned documentation retrieval as a Model Context Protocol server. Gives Claude Code / Cursor / any MCP-aware AI grounded answers from the docs of the exact library version your lockfile pins.

Status: v0.0 — initial scaffold. v0.1 ships PyPI + Smithery registration once the FastMCP server is locally verified.

---

What it is

Claude Code, Cursor, and other AI coding assistants answer library questions from training data. If your project pins react@18.2.0 and the latest is 19.1.0, you get React 19 APIs in your React 18 file — the model has no way to know which version actually matters.

docchat-server is an MCP server that fixes that. Index a library at the version you pin once. Register the server with your MCP host. Now every query to your coding assistant can be grounded in the docs for the exact pinned version, with hard refusal when the docs don't cover the question.

It's the DocChat VS Code extension stripped of its agent + chat UI, exposed as an MCP tool surface instead. The retrieval logic, version-aware routing, and per-library cosine score floors are identical (and identically eval-tuned).

---

Install

pip install docchat-server        # or: uvx --from docchat-server docchat-server

Requires Python 3.11+ and an OPENAI_API_KEY env var (used for query + index-time embeddings). The Qdrant vector store runs embedded — no Docker, no separate server.

---

Use (3 steps)

1. Index the libraries you care about

export OPENAI_API_KEY=sk-...

docchat-server index react 18.2.0
docchat-server index fastapi 0.100.0
docchat-server index vue 3.4.0

Each index takes ~30–60 seconds and a few cents of embedding cost. Stored at ~/.docchat-server/qdrant/.

2. Verify

docchat-server list

Indexed collections:
  - react_18_2_0    (47 chunks)
  - fastapi_0_100_0 (38 chunks)
  - vue_3_4_0       (62 chunks)

Supported libraries: fastapi, react, vue

3. Register with your MCP host

Claude Desktop / Claude Code: add to your MCP config (~/.config/claude/mcp-config.json on Mac/Linux, %APPDATA%\Claude\mcp-config.json on Windows):

{
  "mcpServers": {
    "docchat": {
      "command": "docchat-server",
      "args": ["serve"],
      "env": {
        "OPENAI_API_KEY": "sk-..."
      }
    }
  }
}

Restart your MCP host. The docchat server should appear with two tools: search_docs and list_indexed.

---

The tools

search_docs(library, version, query, api_name?, top_k?)

Retrieves top-K chunks from the indexed docs of the exact pinned version. Returns the chunks with citations, or "No relevant chunks found" if nothing clears the per-library cosine floor (a hard signal to the model that it should refuse rather than guess).

list_indexed()

Returns the collections currently populated locally. Useful as a session-start probe — your assistant can call this once to know what's available before answering anything.

---

Sibling project

The same retrieval engine ships as a VS Code extension on the Marketplace. Source: https://github.com/AshwinUgale/docchat. If you want a chat panel instead of MCP-tool integration, install that.

---

Roadmap

• v0.1 — PyPI publish, Smithery listing, README screenshots from real Claude Code session
• v0.2 — detect_pinned_libraries(workspace_path) tool (parse package.json / pyproject.toml / requirements.txt and report pinned versions to the assistant)
• v0.3 — --repo / --paths flags for arbitrary library indexing (extend beyond the built-in react / fastapi / vue)
• v0.4 — local embeddings via sentence-transformers (drop the OpenAI dependency for the embed step)

---

License

MIT. See LICENSE.