Build One AI Tool Server, Call It From Three Different Agents (MCP Explained)

Short summary

A hands-on tutorial showing how to build one MCP (Model Context Protocol) server in Python that serves image-generation tools to three different AI clients: Claude Code, a Google ADK agent, and a Rust CLI. The server uses FastMCP to expose four tools (generate, edit, edit-local, get-help) backed by Google's Gemini image model via a stateful Interactions API. The key insight is that MCP acts like USB-C for AI tools — write your tool once, call it from any compliant agent.

• •MCP lets one server expose tools to any AI client via JSON over stdio
• •FastMCP turns a decorated Python function into a self-documenting AI tool
• •A single image-generation server serves Claude Code, Google ADK, and Rust CLI simultaneously

Generated with AI, which can make mistakes.