Magellan
Browser based MCP that facilitates operator like tasks when given inputs from the Mistral chat
Core trick
Drives a headless browser through browser-use so Mistral chat can execute web tasks from its inputs.
Description
Browser based MCP that facilitates operator like tasks when given inputs from the Mistral chat
README excerpt
# Project Magellan 🧭 *An instrument of navigation for Mistral AI agents, built for the Mistral AI MCP Hackathon 2025.* [](https://mistral.ai/) [](https://github.com/your-username/magellan) [](LICENSE) --- ## About The Project Le Chat and other Mistral AI-powered assistants are powerful language models, but their capabilities are confined to the digital world of text. They cannot interact with the most significant information interface ever created: the World Wide Web. Popular browser automation libraries like `browser-use` provide the tools for AI agents to control a web browser, but they lack native, first-class support for Mistral's powerful family of models. **Magellan solves this.** Magellan is a custom-built, high-performance **Model Context Protocol (MCP) server** that acts as the missing bridge. It introduces a brand new, purpose-built **Mistral LLM connector** for the `browser-use` library, empowering Mistral agents to navigate, understand, and interact with any website. Our project's goal is to give Le Chat the "eyes" to see the web and the "hands" to use it, all orchestrated through the intelligence of Mistral AI. ## Core Features * **First-Class Mistral Integration:** A custom-built `ChatMistral` class that allows the `browser-use` agent to be pow