gptme ​

gptme is a personal AI agent that lives in your terminal and specializes in running code. It executes Python, shell commands, and browser automation, self-corrects on errors, and chains multiple scripts together — all without leaving the chat. MCP extends its reach by connecting external services the agent can query or modify during execution.

gp
gptme
TRANSPORT
Streamable HTTP ✓
PLATFORM
Windows · macOS · Linux
MCP VIA
Config File

Self-Correcting Execution with External Context ​

gptme's distinguishing feature is its execution loop. When it writes a script that fails, it reads the error output, corrects the code, and retries — all automatically. Adding MCP tools lets the agent gather context before writing code (e.g., fetching a database schema) and validate results after execution (e.g., checking deployment status).

This feedback loop makes gptme well-suited for automation scripts, data analysis pipelines, and DevOps tasks where getting it right on the second try is good enough — and the agent needs live data to do so.

What sets gptme apart:

  • Self-correcting — automatic retry on execution errors
  • Python + shell + browser — three execution environments in one agent
  • Conversation persistence — resume past sessions with full context
  • Pipe-friendly — accepts stdin, works in shell scripts
  • Model-agnostic — Claude, GPT, local Ollama models
  • Web UI — optional browser-based interface alongside the CLI

Connecting Vinkius Cloud ​

1. Create a Token ​

In Vinkius Cloud, navigate to your server → Connection Tokens → Create. Copy the URL.

2. Add to gptme Config ​

Add the MCP server in your gptme configuration:

yaml
mcp_servers:
  vinkius:
    url: "https://edge.vinkius.com/{TOKEN}/mcp"

3. Start a Session ​

bash
gptme

MCP tools load alongside gptme's built-in execution capabilities. The agent calls them as part of its reasoning and execution loop.

FAQ ​

Can gptme run MCP tool calls and Python scripts in the same session? Yes. gptme can query an MCP tool, use the result in a Python script, execute it, and iterate — all within a single conversation turn.

What happens if an MCP tool call fails? gptme treats MCP failures like execution errors: it reads the error, adjusts its approach, and retries with a different strategy.

Does gptme support local models? Yes. It works with Ollama and any OpenAI-compatible local endpoint. MCP functions the same regardless of which model is active.

Is gptme free? Open-source under MIT. Bring your own API key for cloud LLM providers, or use free local models.