source: hugging face blog: adding mcp tools to reachy mini
level: technical
the reachy mini conversation app now supports remote tools called over mcp from public hugging face spaces. you can add a tool like weather or web search with one command, and it runs in the space itself, so no code is downloaded to your machine. built-in tools for robot movement and expressions stay local, while shareable remote tools handle stateless lookups. profiles control which tools are active through a tools.txt file, and remote tools get namespaced ids to avoid collisions.
adding a tool is done with `reachy-mini-conversation-app tool-spaces add <owner/space-name>`. the app validates the space, discovers its tools, and appends them to the active profile's tools.txt. you can install without enabling, list installed spaces, or remove them. two canary tools were created for testing: a search tool and a weather tool. example profiles show how to combine a few expressive built-in tools with remote ones, letting the robot react physically while answering current questions.
prompts matter for how the model uses remote tools. the canary profiles include specific instructions to encourage parallel tool calls when a question mixes weather and search, reducing latency. current limits: only public, mcp-compatible gradio spaces with a standard endpoint are supported; private or non-gradio spaces are not. parallel tool orchestration is not guaranteed and should be handled in code if needed. to publish a tool, create a public gradio space with the mcp endpoint and tag it reachy-mini-tool and mcp.
why it matters: this lets ai and data science practitioners extend robot capabilities with remote, shareable tools without modifying the core app, enabling faster iteration and community-driven tool development.