Me-GPT

A minimal LLM agent CLI for OpenAI, Anthropic, and containerized MCP servers.

Features

• Provider adapters for OpenAI, Anthropic, and generic HTTP (MCP servers)
• Terminal-first CLI with REPL chat and one-off calls
• Simple YAML + environment variable configuration
• In-memory conversation history
• Testable with local mock MCP servers

Installation

For production use, see INSTALL.md for detailed installation instructions including:

• Installing as a standalone executable with pipx
• Installing with pip
• Using the bin/agent shim script
• Troubleshooting Python 3.14 compatibility issues

Quick Start (Development)

poetry install
poetry run agent --help

Configuration

Create a config file at ~/.config/me-gpt/config.yaml:

default_provider: openai
providers:
  openai:
    base_url: https://api.openai.com
    api_key_env: OPENAI_API_KEY
  anthropic:
    base_url: https://api.anthropic.com
    api_key_env: ANTHROPIC_API_KEY
  local_mcp:
    base_url: http://localhost:8080

Set your API keys as environment variables:

export OPENAI_API_KEY="sk-..."
export ANTHROPIC_API_KEY="sk-ant-..."

Usage

Initialize configuration

agent init

One-off completion

agent call --provider openai --prompt "Write a haiku about code"

Interactive chat (REPL)

agent chat --provider openai

Type your messages and press Enter. Use exit or Ctrl+C to quit.

Test providers

agent test

Development

Running the mock MCP server

docker compose -f docker-compose.dev.yml up --build

Or run directly:

cd dev/mock_mcp
poetry run uvicorn server:app --reload --port 8080

Running tests

poetry run pytest

MCP Server HTTP Contract

Your containerized MCP servers should implement:

Request: POST /v1/completions

{
  "model": "gpt-like",
  "input": "prompt text",
  "max_tokens": 256,
  "stream": false
}

Response:

{
  "id": "completion-id",
  "output": "response text",
  "token_usage": {
    "prompt": 10,
    "completion": 20
  }
}