mcp-documentation

This MCP Documentation Server provides a local-first document management system with semantic search and optional AI-powered analysis. Documents are stored in an embedded Orama vector database, enabling hybrid search (full-text plus vector similarity) without requiring external services. A built-in web UI is available by default, offering dashboards, document management, uploading, and context window exploration. You can run the server and interact with tools such as add_document, list_documents, search_all_documents, get_context_window, and more. If you supply a GEMINI_API_KEY, you can enable Gemini-powered AI search to gain deeper contextual insights. Everything runs locally, with data stored under ~./mcp-documentation-server by default (configurable via MCP_BASE_DIR).

Prerequisites:

• Node.js and npm installed on your system
• Internet access to install the MCP server package from npm

Step 1: Install Node.js (if not already installed)

• On macOS: install via Homebrew: brew install node
• On Windows: download and run the Node.js installer from nodejs.org
• On Linux: use your distribution's package manager (e.g., apt, dnf) to install node and npm

Step 2: Install and run the MCP Documentation Server using npx

• The server is published as @andrea9293/mcp-documentation-server. Use the following configuration with npx to install and run automatically:

npx -y @andrea9293/mcp-documentation-server

This will start the MCP server with the default quick-start configuration (documentation channel) and set up the embedded Orama DB for local storage.

Step 3: Optional customization

• If you want to customize the workspace path or enable Gemini AI, set environment variables as shown in the mcp_config section above. You can run with your own environment settings in a shell like:

export MCP_BASE_DIR="/path/to/workspace"
export GEMINI_API_KEY="your-api-key"
export MCP_EMBEDDING_MODEL="Xenova/all-MiniLM-L6-v2"
export START_WEB_UI="true"
export WEB_PORT="3080"
npx -y @andrea9293/mcp-documentation-server

Step 4: Access the web UI

• By default, the built-in web UI starts on http://localhost:3080 (adjust WEB_PORT if you changed it).

Note: Without a GEMINI_API_KEY, only the local embedding-based search tools are available. The server will migrate legacy JSON documents to Orama on first startup automatically.

Tips and considerations:

• Data is stored locally under MCP_BASE_DIR (default: ~/.mcp-documentation-server). Change this to a persistent workspace if needed.
• The web UI can be disabled by setting START_WEB_UI=false. If disabled, you can still interact with the server via its MCP tools programmatically.
• Gemini AI search is optional and requires GEMINI_API_KEY to be set. Without it, semantic search will rely on local embeddings.
• Ensure your environment variables are defined in the same session or via a .env file in the project root for convenience.
• The server supports uploading .txt, .md, and .pdf files to populate the knowledge base; you can also use add_document for manual entries.
• If you’re running behind a firewall or on a non-default port, adjust WEB_PORT accordingly and ensure the port is open.
• The module exposes common MCP tools such as add_document, list_documents, get_document, delete_document, process_uploads, get_uploads_path, list_uploads_files, get_ui_url, search_documents, search_all_documents, and get_context_window.