Multi-Server MCP Documentation Search

A framework for building self-contained, searchable MCP servers from technical documentation. Create independent documentation servers with hybrid vector + keyword search, ready to distribute and deploy.

🎯 Project Overview

This framework enables you to:

1. Build searchable MCP servers from any Markdown/MDX documentation source
2. Search with hybrid semantic (vector) + keyword (FTS) search via Reciprocal Rank Fusion
3. Distribute self-contained servers as standalone repositories or packages
4. Deploy to VS Code, Claude Desktop, or any MCP-compatible host
5. Scale to multiple documentation sources with automated tooling

Core Value Proposition

• 🔍 Hybrid Search: Combines semantic similarity (HNSW) and keyword matching (BM25) intelligently
• 📦 Self-Contained Servers: Each MCP server is fully standalone and distributable
• 🚀 Multi-Format Support: Works with MDX, Markdown, and other documentation formats
• 🎛️ Config-Driven: All paths and parameters controlled via YAML configuration
• 💾 Versioned Data: DuckLake provides reproducible documentation snapshots
• 🔄 Automated Tooling: Scripts for syncing, scaffolding, and building new servers

📐 Multi-Server Architecture

This project supports multiple independent MCP servers, each serving different documentation sources:

/home/james/mcp/
├── servers/                          # Standalone MCP servers
│   ├── mojo-manual-mcp/              # Mojo documentation server
│   │   ├── runtime/                  # Server code + indexed database
│   │   │   ├── mojo_manual_mcp_server.py
│   │   │   ├── search.py
│   │   │   └── mojo_manual_mcp.db
│   │   ├── config/                   # YAML configuration
│   │   │   ├── processing_config.yaml
│   │   │   └── server_config.yaml
│   │   ├── requirements.txt
│   │   └── README.md
│   │
│   └── [future-servers]/             # DuckDB, Python, etc.
│
├── shared/                           # Build-time infrastructure (dev only)
│   ├── preprocessing/                # Document processing pipeline
│   ├── embedding/                    # Embedding generation scripts
│   ├── templates/                    # Templates for new servers
│   └── build/                        # Ephemeral build artifacts
│
├── source-documentation/             # Documentation sources
│   ├── mojo/manual/                  # Mojo docs (MDX files)
│   └── [other-sources]/
│
└── tools/                            # Automation scripts
    ├── sync_documentation.sh         # Sync from upstream repos
    ├── scaffold_new_mcp.sh           # Create new server structure
    └── build_mcp.sh                  # Build server database

Key Design Principles:

• Each server in /servers/{name}/ is completely self-contained and distributable
• Shared build infrastructure in /shared/ is for development only (not packaged with servers)