@piedpiperrh/mcp

MCP (Model Context Protocol) server for Pied Piper — gives Claude and any other MCP client direct access to open-weight LLM inference served by community GPUs.

Tools

| Tool | What it does | | --- | --- | | list_models | Lists model ids currently servable on the network | | chat | Runs a completion on a chosen model (costs credits per request) |

Setup

Mint an API key in the Pied Piper dashboard (Settings → API keys), then:

Claude Code

claude mcp add piedpiper --env PIEDPIPER_API_KEY=lg_pk_your_key -- npx -y @piedpiperrh/mcp

Claude Desktop (claude_desktop_config.json)

{
  "mcpServers": {
    "piedpiper": {
      "command": "npx",
      "args": ["-y", "@piedpiperrh/mcp"],
      "env": { "PIEDPIPER_API_KEY": "lg_pk_your_key" }
    }
  }
}

Then ask Claude things like "use piedpiper to ask llama-3.2-3b what it thinks about X" — Claude calls the network, the request is served by a community host, and the host earns credits.

Environment

| Var | Meaning | | --- | --- | | PIEDPIPER_API_KEY | required — your lg_pk_… key | | PIEDPIPER_BASE_URL | optional — self-hosted/preview deployment origin |