FreeChatMCP

Open-source bridge that lets a supported web-chat AI client control a local Windows computer through MCP, without purchasing a separate LLM API. ChatGPT orchestrates calls in the conversation, while local workers perform explicit tool operations, without needing an extra model API subscription. The web-chat account may still require a paid plan, remain subject to usage limits, and must support MCP/app tool connections.

Web chat is the brain; your computer is the worker. No extra LLM API required.

This is not an API bypass, a way to obtain unlimited AI calls, or a zero-cost guarantee. It uses supported client-side integration and user-authorized tool calls.

v0.2.0 — Browser control and upload

New support for unified browser operations, Windows native IPC, file upload workflows and accompanying tests. This release retains the original MCP tool interfaces and excludes local credentials, runtime state, and browser binaries.

Related MCP server: rvn

Components

• 02_web_mcp: FastAPI/MCP endpoints, OAuth and authorization

• 03_web_worker: deterministic task worker and status reporting

• 06_web_tools: file, browser, shell, Git and optional desktop control tools

• 01_web_frontend: React control panel

• 04_web_canvas and 05_web_memory: local canvas, history and memory

• 07_web_config: shared Python configuration code

• 10_web_deploy: optional Windows services, Edge extension and local deployment scripts

• tests: automated tests

Development setup

Requires Windows, an appropriate Python installation, Node.js/npm for the frontend, and optional third-party runtimes for desktop features.

1. Clone the GitHub repository xiaoruo77/freechatmcp.

2. Run python -m venv .venv in the project directory.

3. Install packages using .venv/Scripts/python.exe -m pip install -e ".[dev]".

4. In 01_web_frontend, run npm install and npm run build.

5. Start run_server.py and run_worker.py using the virtual environment's Python executable.

6. Local UI and MCP endpoint default to 127.0.0.1:48271.

This repository provides source code only. Python/Node environments, third-party Windows-MCP components, cloudflared configuration, browser sessions and installed credentials are not bundled.

Security and configuration

Read .env.example and docs/SECURITY.md. Supply actual values in your process environment; .env.example is not auto-loaded. Default to loopback host, least-privilege workspaces, disabled admin/desktop and shell permissions. Configure and verify any OAuth, public tunnel and browser bridge secrets independently; never publish them. Edge extension config.js is generated locally and ignored by Git.

This is a powerful local automation tool; do not expose it on the internet without correctly configured authentication and access controls.

License

MIT; see LICENSE. Third-party dependencies remain under their own licenses.