Introduction
What and why is tmcp a thing?
If you are ever come close to the world of Agentic AI I'm sure you've come across a certain acronym: MCP.
MCP: Model Context Protocol
As the name suggest Model Context Protocol is...well...a Protocol. Just like HTTP allow browsers (HTTP client) to communicate back and forth with our servers (HTTP server), MCP allows MCP clients to communicate back and forth with MCP servers. To do what? To allow LLMs (the Model in the acronym) to get additional Context (the Context in the acronym) and act on behalf of the user!
The MCP is a way to standardize the communication method between those clients (claude-code, codex, copilot etc) and the servers (what you are probably here to build).
The initiative was launched by Anthropic and quickly gained popularity and now has thousands of users.
If you are interested you can read more here but you are here to learn about tmcp so...
tmcp a modern way to build MCP servers
Most modern platform nowadays standardized on the fetch API to handle their requests/responses. SvelteKit, Next.js, SolidStart, Bun, Cloudflare Workers, Deno, they all handle HTTP calls in the same way: you expose a function that receives a Request instance and return a Response instance from it.
Simple, performant, modern.
There's also a lot of validation libraries out there: Valibot, Arktype, Effect each with it's own strength and as a library author you don't have to pick your favorite: you can use Standard Schema to allow your users to pick their favorite!
This and much more were the reasons that moved us to build tmcp: a simple, composable, modern and flexible SDK to build your MCP server where YOU get to pick your deployment target, you validation library and your preferred transport without all the rest coming as a baggage!
But enough blabblering...you want to see the code!