Geek Out Time: Running FastMCP with DeepSeek on Google Colab

I’ve been reading quite a bit about the MCP recently. It is an open standard designed to solve the “integration tax” of LLMs. Instead of writing customized glue code for every tool or data source, MCP provides a universal interface. Think of it like a Language Server Protocol (LSP) but for AI: it standardizes how a model discovers and interacts with local or remote capabilities.

If you’ve spent any time in the terminal with Claude Code, you know that MCP is basically its nervous system. It’s how the model breaks out of the chat box to actually touch your local files, run a build, or hit an API. Without it, Claude is just a smart guy in a padded room; with it, it’s an engineer with a keyboard.

The Background: Why FastMCP?

For this experiment, I’m using open-source FastMCP ( https://gofastmcp.com/getting-started/welcome ) . If the MCP spec is the underlying JSON-RPC contract, FastMCP is the “FastAPI” of this world. It’s an opinionated, Python-first framework that abstracts the transport logic (STDIO vs. HTTP) and the lifecycle boilerplate.

This is an excerpt — the full article continues on Medium.