docs2db-mcp-server

MCP server for docs2db-api RAG queries. Provides semantic and hybrid search tools for RHEL documentation. Compatible with llama-stack, Goose, Claude Desktop, and other MCP clients.

Features

• Hybrid Search: Combines vector similarity and BM25 keyword search with Reciprocal Rank Fusion
• Cross-Encoder Reranking: Improves result quality using ms-marco-MiniLM-L-6-v2
• MCP Protocol: Standard Model Context Protocol via FastMCP
• SSE Transport: Server-Sent Events for real-time streaming
• Configurable: Environment variables for all settings
• Production Ready: Containerized, non-root user, health checks

Installation

Prerequisites

Install uv:

# macOS/Linux
curl -LsSf https://astral.sh/uv/install.sh | sh

# Or with brew
brew install uv

From PyPI

uv add docs2db-mcp-server

From Source

git clone https://github.com/rhel-lightspeed/docs2db-mcp-server.git
cd docs2db-mcp-server
uv sync
uv pip install -e .

Quick Start

Prerequisites

1. PostgreSQL with pgvector containing docs2db RAG database
2. docs2db to build the RAG database (see docs2db)

Running the Server

# Start the MCP server (uses localhost defaults)
uv run python -m docs2db_mcp

Default configuration connects to postgresql://postgres:postgres@localhost:5432/ragdb

Server will start on http://localhost:8002/sse

Using with llama-stack

Add to your run.yaml:

tool_runtime:
  - provider_id: model-context-protocol
    provider_type: remote::model-context-protocol

registered_resources:
  tool_groups:
    - toolgroup_id: mcp::docs2db-rag
      provider_id: model-context-protocol
      mcp_endpoint:
        uri: http://localhost:8002/sse

Using with Goose

Add to ~/.config/goose/profiles.yaml:

extensions:
  docs2db-rag:
    enabled: true
    type: stdio
    name: docs2db-rag
    cmd: uv
    args: ["run", "python", "-m", "docs2db_mcp"]

Using with Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "docs2db-rag": {
      "command": "uv",
      "args": ["run", "python", "-m", "docs2db_mcp"]
    }
  }
}

Configuration

All configuration via environment variables with DOCS2DB_MCP_ prefix:

MCP Server Settings

Variable	Default	Description
DOCS2DB_MCP_TRANSPORT	sse	Transport type (sse/stdio)
DOCS2DB_MCP_HOST	0.0.0.0	Bind address
DOCS2DB_MCP_PORT	8002	Port number
DOCS2DB_MCP_LOG_LEVEL	INFO	Logging level

Database Settings

Variable	Default	Description
DOCS2DB_MCP_DB_HOST	localhost	PostgreSQL host
DOCS2DB_MCP_DB_PORT	5432	PostgreSQL port
DOCS2DB_MCP_DB_DATABASE	ragdb	Database name
DOCS2DB_MCP_DB_USER	postgres	Database user
DOCS2DB_MCP_DB_PASSWORD	postgres	Database password

RAG Settings

Variable	Default	Description
DOCS2DB_MCP_RAG_SIMILARITY_THRESHOLD	0.7	Minimum similarity score
DOCS2DB_MCP_RAG_MAX_CHUNKS	5	Maximum chunks to return
DOCS2DB_MCP_RAG_ENABLE_RERANKING	true	Enable cross-encoder reranking

Tools

search_documents

Search RHEL documentation using hybrid search (vector + BM25).

Parameters:

• query (string, required): Search query
• max_chunks (integer, optional): Maximum chunks to return (default: 5)
• similarity_threshold (float, optional): Minimum similarity score (default: 0.7)
• enable_reranking (boolean, optional): Enable cross-encoder reranking (default: true)

Returns:

{
  "chunks": [
    {
      "text": "Chunk content...",
      "contextual_text": "LLM-generated context...",
      "similarity": 0.85,
      "source": "path/to/source/file.html",
      "metadata": {...}
    }
  ],
  "query_used": "original query"
}

Docker/Podman

Build Image

podman build -t docs2db-mcp-server .

Run Container

podman run -d \
  --name docs2db-mcp \
  -p 8002:8002 \
  -e DOCS2DB_MCP_DB_HOST=postgres \
  -e DOCS2DB_MCP_DB_PASSWORD=mysecret \
  docs2db-mcp-server

Docker Compose / Podman Compose

services:
  postgres:
    image: pgvector/pgvector:pg17
    environment:
      POSTGRES_PASSWORD: postgres
    volumes:
      - ragdb:/var/lib/postgresql/data
      - ./ragdb_dump.sql:/docker-entrypoint-initdb.d/init.sql

  docs2db-mcp:
    image: docs2db-mcp-server
    ports:
      - "8002:8002"
    environment:
      DOCS2DB_MCP_DB_HOST: postgres
      DOCS2DB_MCP_DB_PASSWORD: postgres
    depends_on:
      - postgres

Development

# Clone repo
git clone https://github.com/rhel-lightspeed/docs2db-mcp-server.git
cd docs2db-mcp-server

# Install dependencies
uv sync
uv pip install -e ".[dev]"

# Run tests
uv run pytest

# Run linters
uv run ruff check .
uv run mypy src/

Architecture

┌─────────────────────────────────────┐
│ MCP Client                          │
│ (llama-stack, Goose, Claude, etc.)  │
└──────────────┬──────────────────────┘
               │ MCP Protocol (SSE/stdio)
               ↓
┌─────────────────────────────────────┐
│ docs2db-mcp-server                  │
│                                     │
│ FastMCP Server                      │
│   └─ search_documents (tool)        │
│         ↓                           │
│   UniversalRAGEngine                │
│   (from docs2db-api)                │
└──────────────┬──────────────────────┘
               │
               ↓
┌─────────────────────────────────────┐
│ PostgreSQL + pgvector               │
│ (docs2db RAG database)              │
└─────────────────────────────────────┘

Related Projects

• docs2db - Build RAG databases from documents
• docs2db-api - Query API for docs2db databases
• linux-mcp-server - MCP server for Linux diagnostics

License

Apache License 2.0 - See LICENSE for details.

Contributing

Contributions welcome! Please open an issue or pull request.

Support

• Issues: https://github.com/rhel-lightspeed/docs2db-mcp-server/issues
• Discussions: https://github.com/rhel-lightspeed/docs2db-mcp-server/discussions