docubridge

MCP server for instant access to your local documentation libraries

docubridge is a Model Context Protocol (MCP) server that gives AI assistants (like Qwen, Claude, etc.) direct access to documentation you specify. Instead of the model guessing or hallucinating API details — it reads the actual docs.

Why?

When working with an LLM in your terminal, the model doesn't know:

• which version of FastAPI you're using

• what your internal project docs say

• any documentation you've added yourself

docubridge solves this by exposing your local Markdown documentation as MCP tools — the model can list, read, and search through them on demand.

Related MCP server: MCP Docs Server

Features

Project Structure

mcp-docs-server/
├── pyproject.toml
├── .env
├── README.md
└── src/
    └── docs_server/
        ├── __init__.py
        ├── main.py
        ├── server.py
        ├── config.py
        └── reader.py

Installation

git clone https://github.com/yourname/docubridge.git
cd docubridge
pip install -e .

Adding Documentation

Clone any documentation that has Markdown sources. For example, FastAPI:

git clone --depth=1 --filter=blob:none --sparse https://github.com/fastapi/fastapi.git
cd fastapi
git sparse-checkout set docs/en/docs

Then move the folder into your docs/ directory:

docs/
└── fastapi/
    ├── index.md
    ├── tutorial/
    └── advanced/

You can add as many libraries as you want — just drop a folder into docs/.

Configuration

Create a .env file in the project root:

DOCS_DIR=./docs

Connecting to Qwen CLI

Add the following to your .qwen/settings.json:

{
  "mcpServers": {
    "docubridge": {
      "command": "docs-server",
      "timeout": 15000
    }
  }
}

Then verify the connection:

qwen mcp list

You should see docubridge in the list.

Usage Examples

Once connected, you can ask your AI assistant:

• "What libraries do you have access to?"

• "Show me all files in the fastapi docs"

• "Find everything about dependency injection in fastapi"

• "Read the content of tutorial/path-params.md"

The model will call the appropriate tool and answer based on the actual documentation.

Requirements

• Python 3.11+

• mcp[cli] >= 1.0.0

• pydantic-settings

License

MIT