Just a few weeks ago, Anthropic announce the Model Context Protocol - a standard for tool integration into AI Apps and Agents.<p>This filled a significant missing link to me - a way for everything to describe tools and make them callable across models, languages, transports etc.<p>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 <a href="https://mcp.run" rel="nofollow">https://mcp.run</a>, the tool registry and control plane to manage MCP "servlets".<p>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.<p>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.<p>Please try it out and give us some feedback.
we opened up a repo of example servlets here:
<a href="https://github.com/dylibso/mcp.run-servlets">https://github.com/dylibso/mcp.run-servlets</a><p>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.