Documentation

The MCP Documentation Server lets you manage and search your documents using AI.

It runs entirely on your local machine and creates a private, searchable knowledge base from your files.

You can add text, markdown, or PDF files and then ask questions about them through an MCP client like Claude Desktop and Cursor.

The server uses AI embeddings to understand the meaning of your text, which results in much smarter search results than simple keyword matching.

Features

• 📄 Document Management – Add, list, retrieve, and delete documents with comprehensive metadata support
• 🔍 Semantic Search – AI-powered search using advanced embeddings for contextual results
• 📁 File Upload Support – Direct processing of .txt, .md, and .pdf files through uploads folder
• 🧩 Smart Text Chunking – Automatic content splitting for improved search accuracy and relevance
• 🗑️ Complete Document Deletion – Clean removal of documents and all associated chunks
• 🌍 Multilingual Support – High-quality embeddings that work across multiple languages
• 💾 Local Data Storage – All information stored locally in your home directory
• ⚡ Zero Configuration Setup – Works immediately without database configuration

Use Cases

• API Documentation Management: Keep multiple API references searchable in one place. Instead of switching between browser tabs or PDF files, search across all your API docs to find specific endpoints, parameters, or implementation examples.
• Code Knowledge Base: Store coding standards, architectural decisions, and team guidelines. When onboarding new developers or making technical decisions, quickly find relevant precedents and established patterns.
• Research Paper Organization: Academic researchers and engineers can upload technical papers and search for specific methodologies, algorithms, or experimental results without manually reading through entire documents.
• Technical Specification Lookup: Maintain searchable access to hardware specifications, protocol definitions, and technical standards. Find specific requirements or compatibility information across multiple technical documents instantly.

How to Use It

1. Installation and Setup

npx @andrea9293/mcp-documentation-server

For permanent installation:

npm install -g @andrea9293/mcp-documentation-server
mcp-documentation-server

2. Add the server to your MCP client configuration file (such as Claude Desktop’s config):

{
  "mcpServers": {
    "documentation": {
      "command": "npx",
      "args": [
        "-y",
        "@andrea9293/mcp-documentation-server"
      ],
      "env": {
        "MCP_EMBEDDING_MODEL": "Xenova/all-MiniLM-L6-v2"
      }
    }
  }
}

3. Available Tools and Methods

Document Management Tools:

• add_document – Creates new documents with title, content, and metadata
• Parameters: title (string), content (string), metadata (object)
• Returns: Document ID and confirmation
• list_documents – Retrieves all documents with their metadata
• Returns: Array of document objects with IDs, titles, and metadata
• get_document – Fetches complete document content by ID
• Parameters: id (string)
• Returns: Full document with all chunks
• delete_document – Removes document and all associated chunks
• Parameters: id (string)
• Returns: Deletion confirmation

Search Tools:

• search_documents – Performs semantic search within specific documents
• Parameters: document_id (string), query (string), limit (number)
• Returns: Ranked search results with relevance scores

File Processing Tools:

• get_uploads_path – Returns the uploads directory path
• Returns: Full path to uploads folder (~/.mcp-documentation-server/uploads/)
• list_uploads_files – Shows all files in uploads directory
• Returns: Array of file names and types
• process_uploads – Converts uploaded files into searchable documents
• Processes: .txt, .md, and .pdf files
• Returns: Processing results and new document IDs

4. Configuration Options

Embedding Models: Set via MCP_EMBEDDING_MODEL environment variable:

• Xenova/all-MiniLM-L6-v2 (default) – Fast processing, good quality results
• Xenova/paraphrase-multilingual-mpnet-base-v2 – Best quality, supports multiple languages

Data Storage: All data is stored locally in your home directory:

~/.mcp-documentation-server/
├── data/      # Document storage in JSON format
└── uploads/   # Files awaiting processing

5. Step-by-Step Usage Workflow

Add Documents Directly:

{
  "tool": "add_document",
  "arguments": {
    "title": "React Hooks Guide",
    "content": "React Hooks allow you to use state and other React features...",
    "metadata": {
      "category": "frontend",
      "framework": "react",
      "tags": ["hooks", "state-management"]
    }
  }
}

Upload Files for Processing:

• Get uploads path: get_uploads_path
• Copy your .txt, .md, or .pdf files to the uploads directory
• Run process_uploads to convert files into searchable documents

Search Your Documents:

{
  "tool": "search_documents",
  "arguments": {
    "document_id": "doc-abc123",
    "query": "useState hook implementation",
    "limit": 10
  }
}

FAQs

Q: Why does the server seem to hang the first time I run it?
A: The first time you start the server, it needs to download the AI embedding model. The recommended model is over 400MB, so it can take a few minutes depending on your internet connection. Just let it run in the background; it will become responsive once the download is complete.

Q: What happens if I change the embedding model after adding documents?
A: The embeddings created by different models are not compatible. If you switch models, your existing documents will no longer be found in searches. You must delete all documents and re-add them to be indexed with the new model.

Q: Does this server extract text from scanned PDFs or images?
A: No, it does not perform Optical Character Recognition (OCR). It can only extract text that is already digitally embedded in the PDF file.

Q: Is my data sent to any external services?
A: No. The entire server, including the documents and the AI models, runs on your local machine. Your data is stored in the ~/.mcp-documentation-server/ directory and is never sent over the internet.