XiaoPuOuO/openchatx-mcp

Turn ChatGPT into a local agent runtime. Operate your computer, use local tools, discover MCP servers, and delegate work to your own models through one MCP connection.

What it solves

OpenChatX provides a comprehensive, local runtime environment that extends ChatGPT's capabilities. It allows users to build and manage complex agent workflows—such as file manipulation, shell command execution, and computer control—without needing to build a custom agent runtime from scratch or reverse-engineer ChatGPT's internal systems.

How it works

It leverages OpenAI's official Model Context Protocol (MCP) and Secure MCP Tunnels. OpenChatX runs a local MCP server that connects to ChatGPT through an official, secure tunnel. When ChatGPT needs to perform a task, it sends a request through this tunnel to your local machine, where OpenChatX executes the action using your local permissions. This setup ensures that ChatGPT remains the primary planner while OpenChatX acts as the robust, local tool-execution layer.

Who it’s for

This project is for power users and developers who want to turn ChatGPT into a more capable, autonomous agent that can interact with their local files, terminal, and custom tools while maintaining a secure, official connection.

Highlights

  • Official Integration: Uses documented OpenAI Secure MCP Tunnels rather than unofficial hacks or browser cookie scraping.
  • Cross-Session Persistence: Features a summarization system that allows long-running agent tasks to survive across multiple ChatGPT conversations.
  • Extensible Tooling: Includes support for custom TypeScript Toolboxes, portable Skills, and persistent Rules (compatible with Cursor and Claude formats).
  • Project Management: Allows binding specific work to local folders and managing subagents for delegated tasks.
  • Desktop Convenience: Provides pre-built desktop applications for macOS and Windows that handle the underlying infrastructure automatically.

Related

  • Project
    SleepySoft/IntelligenceIntegrationSystem

    An OSINT news intelligence system that crawls public news from multiple sources, uses AI to analyze and score the value of each article, and provides a web interface for browsing and searching intelligence. It aims to filter out noise and highlight important information.

  • Project
    MAA-AI/MaaMCP

    An MCP server that enables AI assistants to automate Android devices and Windows desktop applications using OCR and background control.

  • Project
    devforth/adminforth

    AdminForth is a free, MIT-licensed TypeScript/Vue 3 framework for building business admin panels that AI agents can both build and operate. Data stays in your own database; agents connect via MCP; permissions and an audit log (naming person and agent) apply to every request; visual screens are optional. Supports Postgres, MySQL, SQLite, MongoDB, ClickHouse, with plugins for security, import/export, dashboards, and more. Requires Node.js 22.12+.

  • Project
    scaleapi/agentenv-framework

    AgentEnv Framework is a Python SDK/CLI for building, deploying, and scoring agentic environments (Docker or sandbox containers) that communicate via a standard protocol, enabling easy evaluation of LLM‑style agents in defined tasks.