Model Context Protocol (MCP) Server

** NOTE: This project is no longer being maintained, as issues with model memory for smaller models means my preferred method for MCP servers is as individual containerized server providers rather than a monolithic routing provider. Code is being kept up in case others still want to try to go down this route.

Model Context Protocol (MCP) Server

A modular server that implements the Model Context Protocol standard, providing tools for GitHub, GitLab, Google Maps, Memory storage, and Puppeteer web automation.

Architecture

The MCP server is built with a modular architecture, where each tool is implemented as a separate module. The server provides a unified gateway that routes requests to the appropriate tool.

Features

• MCP Gateway: A unified endpoint for all tool requests following the MCP standard
• MCP Manifest: An endpoint that describes all available tools and their capabilities
• Direct Tool Access: Each tool can be accessed directly via its own API endpoints
• Modular Design: Easy to add or remove tools as needed

Included Tools

1. GitHub Tool: Interact with GitHub repositories, issues, and search
2. GitLab Tool: Interact with GitLab projects, issues, and pipelines
3. Google Maps Tool: Geocoding, directions, and places search
4. Memory Tool: Store and retrieve data persistently
5. Puppeteer Tool: Take screenshots, generate PDFs, and extract content from websites

Getting Started

Prerequisites

• Python 3.8 or higher
• Node.js 14 or higher
• A Red Hat-based Linux distribution (RHEL, CentOS, Fedora) or any Linux/macOS system

Installation

1. Clone this repository:

   git clone https://github.com/yourusername/mcp-server.git
   cd mcp-server
   

2. Install Python dependencies:

   pip install -r requirements.txt
   

3. Install Node.js dependencies:

   npm install
   

4. Create a .env file with your configuration:

   SECRET_KEY=your-secret-key
   DEBUG=False
   
   # GitHub configuration
   GITHUB_TOKEN=your-github-token
   
   # GitLab configuration
   GITLAB_TOKEN=your-gitlab-token
   
   # Google Maps configuration
   GMAPS_API_KEY=your-google-maps-api-key
   
   # Memory configuration
   MEMORY_DB_URI=sqlite:///memory.db
   
   # Puppeteer configuration
   PUPPETEER_HEADLESS=true
   CHROME_PATH=/usr/bin/chromium-browser
   

5. Start the server:

   python app.py
   

Containerized Deployment

You can run the server using either Docker or Podman (Red Hat's container engine).

Docker Deployment

If you already have Docker and docker-compose installed:

1. Build the Docker image:

   docker build -t mcp-server .
   

2. Run the container:

   docker run -p 5000:5000 --env-file .env mcp-server
   

3. Alternatively, use docker-compose:

   Create a docker-compose.yml file:

   version: '3'
   services:
     mcp-server:
       build: .
       ports:
         - "5000:5000"
       volumes:
         - ./data:/app/data
       env_file:
         - .env
       restart: unless-stopped
   

   Then run:

   docker-compose up -d
   

Podman Deployment

For Red Hat based systems (RHEL, CentOS, Fedora) using Podman:

1. Build the container image:

   podman build -t mcp-server .
   

2. Run the container:

   podman run -p 5000:5000 --env-file .env mcp-server
   

3. If you need persistent storage:

   mkdir -p ./data
   podman run -p 5000:5000 --env-file .env -v ./data:/app/data:Z mcp-server
   

   Note: The :Z suffix is important for SELinux-enabled systems.

4. Using Podman Compose (if installed):

   # Install podman-compose if needed
   pip install podman-compose
   
   # Use the same docker-compose.yml file as above
   podman-compose up -d
   

Using the MCP Server

MCP Gateway

The MCP Gateway is the main endpoint for accessing all tools using the MCP standard.

Endpoint: POST /mcp/gateway

Request format:

{
  "tool": "github",
  "action": "listRepos",
  "parameters": {
    "username": "octocat"
  }
}

Response format:

{
  "tool": "github",
  "action": "listRepos",
  "status": "success",
  "result": [
    {
      "id": 1296269,
      "name": "Hello-World",
      "full_name": "octocat/Hello-World",
      "owner": {
        "login": "octocat",
        "id": 1
      },
      ...
    }
  ]
}

MCP Manifest

The MCP Manifest describes all available tools and their capabilities.

Endpoint: GET /mcp/manifest

Response format:

{
  "manifestVersion": "1.0",
  "tools": {
    "github": {
      "actions": {
        "listRepos": {
          "description": "List repositories for a user or organization",
          "parameters": {
            "username": {
              "type": "string",
              "description": "GitHub username or organization name"
            }
          },
          "returns": {
            "type": "array",
            "description": "List of repository objects"
          }
        },
        ...
      }
    },
    ...
  }
}

Direct Tool Access

Each tool can also be accessed directly via its own API endpoints:

• GitHub: /tool/github/...
• GitLab: /tool/gitlab/...
• Google Maps: /tool/gmaps/...
• Memory: /tool/memory/...
• Puppeteer: /tool/puppeteer/...

See the API documentation for each tool for details on the available endpoints.

Tool Documentation

GitHub Tool

The GitHub tool provides access to the GitHub API for repositories, issues, and search.

Actions:

• listRepos: List repositories for a user or organization
• getRepo: Get details for a specific repository
• searchRepos: Search for repositories
• getIssues: Get issues for a repository
• createIssue: Create a new issue in a repository

GitLab Tool

The GitLab tool provides access to the GitLab API for projects, issues, and pipelines.

Actions:

• listProjects: List all projects accessible by the authenticated user
• getProject: Get details for a specific project
• searchProjects: Search for projects on GitLab
• getIssues: Get issues for a project
• createIssue: Create a new issue in a project
• getPipelines: Get pipelines for a project

Google Maps Tool

The Google Maps tool provides access to the Google Maps API for geocoding, directions, and places search.

Actions:

• geocode: Convert an address to geographic coordinates
• reverseGeocode: Convert geographic coordinates to an address
• getDirections: Get directions between two locations
• searchPlaces: Search for places using the Google Places API
• getPlaceDetails: Get details for a specific place

Memory Tool

The Memory tool provides a persistent key-value store for storing and retrieving data.

Actions:

• get: Get a memory item by key
• set: Create or update a memory item
• delete: Delete a memory item by key
• list: List all memory items, with optional filtering
• search: Search memory items by value

Puppeteer Tool

The Puppeteer tool provides web automation capabilities for taking screenshots, generating PDFs, and extracting content from websites.

Actions:

• screenshot: Take a screenshot of a webpage
• pdf: Generate a PDF of a webpage
• extract: Extract content from a webpage

Contributing

Contributions are welcome! Here's how you can extend the MCP server:

Adding a New Tool

1. Create a new file in the tools directory, e.g., tools/newtool_tool.py
2. Implement the tool with actions following the same pattern as existing tools
3. Add the tool to the manifest in app.py
4. Register the tool's blueprint in tools/__init__.py

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgements

• Model Context Protocol for the standard specification
• Flask for the web framework
• Puppeteer for web automation