mcp.run: writing your own Wasm extensions for AI and beyond! [Workshop]

Edoardo Vacchi - Dylibso

Months ago, Anthropic announced the Model Context Protocol (MCP), a specification to connect AI models to different data sources and tools.

Only days later, we shipped mcp.run with our own spin on MCP servers: an “app store” for portable Wasm servlets.

An “MCP server” runs locally and never leaves your machine, it provides integration and context to the AI model so that it can answer more questions, giving you control over what data it can access. However, running multiple integration servers creates a maintenance burden for both the users who have to install and keep them up to date, and the developers who will now have to support a variety of platforms.

What if there was a technology that allowed you to host many self-contained guests in a local, safe, sandboxed host environment? Oh, wait!

In this workshop:

• You will build your first Wasm extension to an AI assistant!
• You will learn how LLMs interact with our system
• we will show you how to build your own mcp.run client, and integrate it with your favorite LLM!

You will walk out knowing more about writing Wasm extension for a compelling use case, using tools such as:

• the Extism open source plug-in framework
• the XTP Wasm deployment and delivery service
• the extism and xtp command-line interfaces for code-gen and testing to get you up and running quickly