What is Mcp Pack

MCP Pack is a tool designed for creating and managing documentation databases that are sourced from GitHub repositories. It allows users to facilitate semantic search functionalities over repository documentation by utilizing a Qdrant server for indexing and searching purposes.

Use cases

MCP Pack can be used by developers and teams to create searchable documentation databases for their projects, streamline information retrieval from code repositories, and enhance collaborative efforts in understanding repositories by summarizing Jupyter notebooks and including additional documentation formats such as RST.

How to use

To use MCP Pack, start by running a Qdrant server with Docker and create a documentation database by executing the command ‘mcp_pack create_db’ followed by the repository URL. It also supports various options like including Jupyter notebooks, setting the output directory, and defining a module name. After database creation, you can start a server for querying the documentation with ‘mcp_pack create_server’.

Key features

Key features of MCP Pack include easy integration with GitHub repositories, the ability to create and manage multiple documentation collections, support for including Jupyter notebooks and RST files, options for verbose output, and a query server that can use different transport methods. Additionally, it offers options to clean and list database collections.

Where to use

MCP Pack is particularly useful in software development environments where documentation is essential for onboarding, collaboration, and maintenance. It can be implemented in organizations that rely on GitHub for version control and need to provide easy access to their project documentation for developers, researchers, and other stakeholders.

MCP Pack

A tool for creating and managing documentation databases from GitHub repositories.

Quickstart to create a database

Start the Qdrant server:

docker compose up

With a Qdrant server running:

uvx mcp_pack create_db https://github.com/user/repo

To start a server for querying the documentation:

mcp_pack create_server --module-name your_module_name

see the examples/ folder for how to setup the MCP server.

Installation

# Install from pip
pip install mcp_pack

Prerequisites

• Qdrant server running (by default at http://localhost:6333)
• GitHub token (optional, but recommended to avoid rate limits)
• OpenAI API key (optional, for summarizing Jupyter notebooks)

Usage

See example/ folder.

Create a documentation database

# Basic usage
mcp_pack create_db https://github.com/user/repo

# With @ prefix syntax
mcp_pack create_db @https://github.com/user/repo

# With additional options
mcp_pack create_db @https://github.com/user/repo \
    --output-dir ./output \
    --verbose \
    --include-notebooks \
    --include-rst \
    --github-token YOUR_GITHUB_TOKEN \
    --openai-api_key YOUR_OPENAI_API_KEY

Clean the database

# Delete all collections
mcp_pack clean_db

# Delete a specific collection
mcp_pack clean_db --collection repo-name

List Database Collections

To list all collections in the Qdrant database, use the following command:

# Use default Qdrant
mcp_pack list_db

# Use Custom Qdrant server URL 
mcp_pack list_db --qdrant-url http://localhost:6333

This will display all the collections currently stored in the Qdrant database.

Create and Run a Query Server

To create and run a server for querying module documentation:

# Basic usage with default settings
mcp_pack create_server --module-name your_module_name

# With custom settings
mcp_pack create_server --module-name your_module_name \
    --transport sse \
    --port 8080 \
    --qdrant-url http://your-qdrant-server:6333 \
    --encoder-model all-MiniLM-L6-v2 \
    --collection-name custom_collection

This will start a server that provides semantic search capabilities over your module’s documentation.

Environment Variables

You can set environment variables instead of passing command-line arguments:

# Create a .env file
GITHUB_TOKEN=your_github_token
OPENAI_API_KEY=your_openai_api_key

Options

create_db

• repo_url: GitHub repository URL (can be prefixed with @)
• --output-dir, -o: Directory to save JSONL output
• --verbose, -v: Verbose output
• --include-notebooks: Include Jupyter notebooks
• --include-rst: Include RST files
• --exclude-tests: Exclude test files and directories
• --module-name: Name of the module (defaults to repository name)
• --db-path: Path to store the database
• --qdrant-url: Qdrant server URL (default: http://localhost:6333)
• --github-token: GitHub personal access token
• --openai-api-key: OpenAI API key

clean_db

• --qdrant-url: Qdrant server URL (default: http://localhost:6333)
• --collection: Specific collection to delete (optional)

list_db

• --qdrant-url: Qdrant server URL (default: http://localhost:6333)

create_server

• --module-name: Name of the module to query (required)
• --qdrant-url: Qdrant server URL (default: http://localhost:6333)
• --encoder-model: SentenceTransformer model to use (default: all-MiniLM-L6-v2)
• --collection-name: Name of the Qdrant collection (defaults to module_name)
• --transport: Transport method for the MCP server (default: stdio, choices: stdio, sse)
• --port: Port number for the MCP server (default: 8000)

Global Options

• --version: Show version information

Additional info

> python -m mcp_pack.create_db --help 
Create documentation database for a GitHub repository

positional arguments:
  repo_url              GitHub repository URL

options:
  -h, --help            show this help message and exit
  --output-dir OUTPUT_DIR, -o OUTPUT_DIR
                        Directory to save JSONL output
  --verbose, -v         Verbose output
  --include-notebooks   Include Jupyter notebooks
  --include-rst         Include rst files
  --db-path DB_PATH     Path to store the database
  --qdrant-url QDRANT_URL
                        Qdrant server URL
  --github-token GITHUB_TOKEN
                        GitHub personal access token
  --openai-api-key OPENAI_API_KEY
                        OpenAI API key

python -m mcp_pack.clean_db --help
Clean Qdrant database collections

options:
  -h, --help            show this help message and exit
  --qdrant-url QDRANT_URL
                        Qdrant server URL
  --collection COLLECTION
                        Specific collection to delete (optional, if not provided, all collections will be deleted)

Resources

• Qdrant quickstart
• Qdrant API