Overview

Google, IBM & Meta Certificates – 40% Off

One Coursera Plus subscription covers most Professional Certificates on Coursera.

An agent that only generates text can't do real work. This course connects your agents to databases, APIs, and external services using custom tools and Model Context Protocol (MCP) servers, with the security controls to meet production demands. You'll define typed tool schemas that agents discover on their own, wire up local and remote MCP servers, enforce least-privilege permissions, and handle failures gracefully. By the end, you'll be able to: - Build custom tools with typed schemas that agents invoke autonomously - Configure MCP servers for external data and service access - Implement least-privilege permission controls for tool calls - Build error-handling patterns that keep workflows running Who this is for: Developers who want to extend their agents with custom tools, external data sources via MCP servers, and production-grade error handling. Tools include the GitHub Copilot SDK (Python and Node.js/TypeScript), MCP servers, the GitHub Copilot CLI, VS Code, and Git/GitHub. No prior experience with AI agents or the GitHub Copilot SDK is required.

Syllabus

Defining & Registering Custom Tools
- Define and register custom tools that Copilot agents can autonomously discover and invoke during task execution. You'll learn to create tools with typed parameter schemas and return types, register them within sessions, and verify that agents correctly select and invoke your tools based on task requirements.
Integrating MCP Servers for External Data
- Configure Model Context Protocol (MCP) servers to give your agents access to external data sources and services. MCP is the SDK's primary extensibility mechanism for connecting agents to the outside world, databases, APIs, file systems, and third-party services. You'll learn to set up local and remote MCP servers and integrate them into Copilot sessions.
Implementing Permission Controls for Tools
- Implement permission controls that enforce least-privilege access for agent tool invocations. You'll learn to design approval workflows for sensitive operations, classify tools by risk level, and build permission handlers that give humans oversight of high-stakes agent actions. A critical production safety requirement for agentic systems
Error Handling & Tool Reliability
- Implement robust error handling for tool invocations that keeps agent workflows running when tools fail. You'll learn to design retry strategies, implement fallback tools, and build graceful degradation patterns that allow agents to continue producing useful results even when their tools encounter errors.
Project Module: Tool-Integrated Agent with MCP Connectivity
- Build a tool-integrated agent that combines custom tools, Model Context Protocol (MCP) server connectivity, permission controls, and error handling into a functioning data research assistant. This project integrates skills into a portfolio-ready deliverable that demonstrates production-grade tool management.