Mcp.run – Universal Tools for AI

Just a few weeks ago, Anthropic announce the Model Context Protocol - a standard for tool integration into AI Apps and Agents.

This filled a significant missing link to me - a way for everything to describe tools and make them callable across models, languages, transports etc.

But, the issue of tool code portability and security was left unaddressed, until today! We're excited to ship an early version of `mcpx`, the extensible & dynamically re-programmable MCP Server, as well as https://mcp.run, the tool registry and control plane to manage MCP "servlets".

All servlets are WebAssembly modules, and thus can be called from any of your AI applications in any language - they run locally, on your desktop or in the browser, in a cloud server, or IoT devices & everywhere in between.

AI applications will be deployed on every platform & in every shape and size. I think it is imperative that the tools these AI apps need can also be deployed there too. That's the power of mcp.run & mcpx.

Please try it out and give us some feedback.

we opened up a repo of example servlets here: https://github.com/dylibso/mcp.run-servlets

These can be written in virtually any language, so you can bring advance libraries from your language of choice into your agents and LLMs - no longer restricted to whatever the language implementation of the Agent SDK you're using.