MCP (Model Context Protocol) Integration

OpenPango ships with a built-in MCP server that makes every installed skill available as a tool in any MCP-compatible client — Claude Desktop, Cursor, Windsurf, and more.

What is MCP?

The Model Context Protocol is the emerging standard for connecting AI models to external tools. Instead of writing custom integrations for each LLM client, MCP provides a single protocol that all clients understand.

OpenPango's MCP server auto-discovers all installed skills and registers them as MCP tools. When you add a new skill, it's automatically available in every connected client.

Quick Start

1. Register with Claude Desktop

python3 -m skills.mcp.mcp_server --register-claude

This prints the config snippet to add to: ~/Library/Application Support/Claude/claude_desktop_config.json

2. List Available Tools

python3 -m skills.mcp.mcp_server --list-tools

Currently discovers 28 tools from all installed OpenPango skills, plus two built-in tools:

• openpango_pool_stats — Live mining pool statistics
• openpango_list_skills — List all installed skills

3. Manual Config

Add this to your MCP client's configuration:

{
  "mcpServers": {
    "openpango": {
      "command": "python3",
      "args": ["-m", "skills.mcp.mcp_server"],
      "cwd": "/path/to/openpango-skills"
    }
  }
}

Architecture

The MCP server uses stdio transport (JSON-RPC over stdin/stdout):

MCP Client ←→ stdin/stdout ←→ mcp_server.py ←→ OpenPango Skills

Supported methods:

• initialize — Protocol handshake
• tools/list — List all available tools
• tools/call — Execute a tool
• resources/list — List available resources

Configuration

Create mcp_config.json in your project root to customize behavior:

{
  "server": {
    "name": "openpango-skills",
    "version": "1.0.0"
  },
  "skills": {
    "expose_all": true,
    "allowlist": ["browser", "memory", "mining", "mcp"]
  },
  "auth": {
    "require_token": false
  }
}

Set expose_all to false and use allowlist to restrict which skills are exposed.

MCP Client

OpenPango also includes an MCP client for consuming external MCP servers:

from skills.mcp.mcp_client import MCPClient

with MCPClient("python3", ["-m", "external_server"]) as client:
    tools = client.list_tools()
    result = client.call_tool("web_search", {"query": "OpenPango"})