Reachy Mini can now search the web and check weather with one command

Pollen Robotics just made Reachy Mini a lot more useful outside the lab. The robot can now pull in live web search results and weather data during conversations, capabilities that were previously impossible because every tool had to run as local Python code. The team solved this by introducing remote tools hosted as MCP Spaces, which install from the Hugging Face Hub with a single command.

The architecture is clean. Built-in tools like move_head and play_emotion stay local because they talk to hardware. Remote tools handle stateless lookups—search, weather, anything you’d want to iterate on without touching the app itself. Each tool gets discovered when you install the Space, and a profile’s tools.txt file acts as the gatekeeper. If a tool name isn’t in that file, the model can’t call it. The namespacing scheme uses double underscores to keep pollen_robotics_reachy_mini_weather_tool__get_day_brief from colliding with anything built-in.

They started with two canary tools to exercise the whole flow. The search tool and weather tool are small by design—just enough to validate that installation, discovery, and realtime calls all work the same way local tools do. What’s smart is how they stack: add both Spaces, and the robot can search the web and check the weather in the same conversation. The canary_web_search_weather profile demonstrates exactly that.

Anyone can publish a compatible Space, which opens the door for a community tool ecosystem. The constraints are minimal—it’s a good fit for anything stateless. I’m curious how they’ll handle tools that need authentication or persistent sessions, since the current design seems optimized for public lookups. But for a first iteration, the implementation is refreshingly straightforward. No overengineering, just a CLI command, a namespacing convention, and profiles that stay in control of what the model can access.