tupac

A GPT wrapper with MCP support. It's a thin layer around the OpenAI responses API with functions being specified as MCP config.

You can write a simple "LLM app" very quickly, by specifying MCP config and a system prompt.

MCP functionality supported:

• ✅ tools
• ✅ resources — but only as far as they're being returned from tools. No listing or fetching.

Nothing else. It's what I consider to be an absolute bare-bones MCP app.

Usage: LLM app

uvx tupac configs/web-search.json "When are we getting to Mars?"

Configuration files may contain ${VARNAME} placeholders which are expanded from the environment before parsing. Environment variables can also be loaded from a .env file via python-dotenv. See configs/web-search.json for an example using ${EXA_API_KEY}.

Configuration format follows the standard MCP schema:

{
  "instructions": "Use search to answer questions.",
  "model": "o3",
  "mcpServers": {
    "exa": {
      "type": "url",
      "url": "https://mcp.exa.ai/mcp?exaApiKey=${EXA_API_KEY}"
    }
  }
}

You can use that ${EXA_API_KEY} syntax to reference environment variables. It does load .env files.