Thirdbrain MCP OpenAI Agent — Communication skill for Claude Code

ThirdBrAIn MCP OpenAI agent

This agent created for the oTTomator hackathon makes the LiveAgentStudio work in similar ways to Claude Desktop but for OpenAI compatible models.

This agent makes mcp tools accessible to openai compatible llms (chat completions). This enables the use of MCP servers across a much wider range of models such as:

• evidently OpenAI such as GPT-4o-mini, GPT-4o
• deepseek-chat (currently v3) , at this point deepseek-r1 does not support function calling.
• ollama with deepseek variants (not tested yet).

MCP makes it possible to centralize resources (RAG), prompts and tools and use the same code across platforms. It is an extreme DRY concept for distributed agentic systems that hides the implementation details. For more information see https://modelcontextprotocol.org

Demo

Planning for skiweekend

The agent has built-in python code for weather forecast service for US. I asked where to go skiing around Denver next weekend and did some follow-up calls to get exact weather conditions and a comparison between ski resorts demonstrating iterative tool calling until the weather forecast for each ski resort is obtained and recommendations made. 

Installation

If you are unfamiliar with the hackathon agents, follow the detailed installation instructions provided for the hackathon. [Live Agent Studio Hackaton Readme](LiveAgentStudio_README.md)

Specifics for the MCP OpenAI agent

This agent has uses or has been tested with:

• Python 3.12
• pip
• Supabase
• OpenAI API Key, or Deepseek API key, or no key if using Ollama
• Appropriate runtime environment for the mcp server commands you want to use such as
  • Docker
  • Node / npx
  • Uv / uvx
  • Python3

Note: Anthropic and Docker have published servers that are easy to install. PulseMCP website has the most comprehensive collection of MCP servers.

.env

Rename the env example file `mv .env.example .env`

Use your f