My Skill Lab
Artificial IntelligenceAdvanced

Tool Use and MCP: Connecting LLMs to Real Systems

Function calling, Model Context Protocol servers and safe integrations.

Taught by Rohan Maharjan · ML Engineer

What you'll learn

  • Implement a function-calling loop that runs tools and returns results to the model
  • Design tool names, descriptions and schemas that models call correctly
  • Build a Python MCP server exposing tools, resources and prompts
  • Run servers over stdio or Streamable HTTP and connect them to Claude Desktop and Claude Code
  • Secure remote servers with OAuth 2.1, scoped tokens and human approval
  • Test servers with MCP Inspector and automated client scripts

About this course

Tool Use and MCP is an advanced, hands-on course for software and ML engineers who want large language models to act on real systems safely. You will start with LLM function calling in the Claude API and learn to design tool schemas that models use correctly, then move to the Model Context Protocol (MCP), the open standard for connecting AI assistants to data, APIs and developer tools. You will build an MCP server with the official Python SDK v2, run it over stdio and Streamable HTTP, connect it to Claude Desktop and Claude Code, secure it with OAuth 2.1 and scoped permissions, defend against prompt injection and tool poisoning, and test it with MCP Inspector. You should be comfortable with Python, HTTP APIs and JSON before starting.

Course content

7 lessons
  1. 1From Function Calling to MCPHow models call tools, what the tool loop looks like, and why MCP standardised it.Preview
  2. 2Designing Tools Models Use CorrectlyNames, descriptions, schemas, errors and granularity choices that make tools reliable.
  3. 3MCP Architecture: Hosts, Clients and ServersThe roles, lifecycle, JSON-RPC messages and primitives that make up the Model Context Protocol.
  4. 4Building an MCP Server in PythonUse the official Python SDK v2 and MCPServer to expose tools, resources and prompts from a real service.
  5. 5Transports, Clients and Remote DeploymentRun servers over stdio or Streamable HTTP and connect them to Claude Desktop, Claude Code and other hosts.
  6. 6Authorisation and PermissionsProtect remote MCP servers with OAuth 2.1, scoped tokens and least-privilege tool design.
  7. 7Security, Testing and Production ReadinessDefend against prompt injection and tool poisoning, then test servers with MCP Inspector and client scripts.