Mcpixir is a library for connecting LLMs (Language Learning Models) to MCP (Machine Control Protocol) servers.

This library enables AI agents to access external tools like web browsing, file operations, and specific applications through a unified interface. It's designed to connect any LLM to various tools through MCP servers.

Main Features

• LLM Integration: Connect any LLM with tool-calling capabilities (OpenAI, Anthropic, etc.)
• Multiple Connection Types: Support for HTTP, WebSocket, and stdio connections
• Server Management: Use multiple MCP servers simultaneously, with dynamic server selection
• Comprehensive Tool Access: Interact with web browsers, file systems, and specialized applications
• Easy Configuration: Simple configuration options via code or JSON files

Getting Started

Create a client and agent:

# Create a client with a configuration
client = Mcpixir.new_client(%{
  mcpServers: %{
    browser: %{
      command: "npx",
      args: ["@playwright/mcp@latest"]
    }
  }
})

# Create an agent with an LLM
{:ok, agent} = Mcpixir.new_agent(%{
  llm: %{
    provider: :openai,
    model: "gpt-4o"
  },
  client: client
})

# Run a query
{:ok, result, updated_agent} = Mcpixir.run(agent, "Search for Elixir libraries on GitHub")

Examples

The library includes several Mix tasks for demonstrating various use cases:

# Run the browser example
mix mcp.browser

# Run the Airbnb search example with Anthropic
mix mcp.airbnb --provider=anthropic

See the README for more detailed examples and the full API reference.