← back

Real world MCPs in GitHub Copilot Agent Mode — Jon Peck, Microsoft

5.0K views · Jul 19, 2025 · 14:27 min · Watch on YouTube ↗
Takeaway

Copilot Agent Mode + MCPs makes IDE agents capable of complex multi-tool tasks while keeping the developer in the permission loop.

Summary

  • GitHub Copilot Agent Mode does end-to-end tasks (green-field apps, deep refactors) by reading a thorough README, asking for terminal/tool permissions iteratively, then iterating with the user.
  • MCP (Model Context Protocol) extends agent mode by letting Copilot connect to local DBs, web APIs, or remote servers via JSON config in VS Code settings.
  • Live example: PostgreSQL MCP (read-only) lets Copilot pull schema and seed test data into a mock.json for test generation — safe because no mutation possible.
  • One-click install from github.com/modelcontextprotocol/servers; users must enable each tool and prompts should explicitly mention 'use the Postgres MCP' for reliable selection.
  • Best practice: complete READMEs, copilot-instructions.md, and prompt files materially improve first-try output.
mcpcopilotagents
Original description
As developers, we don't spend most of our time vibe-coding prototypes. More often, we're adding features, squashing bugs, and building tests for existing apps across a wide variety of services and technologies. Come learn how MCPs help GitHub Copilot to untangle real engineering problems. By allowing agent mode to securely work with data sources, testing tools, infrastructure providers, and even core DevOps tooling -- we can go beyond the hype, and solve the actual engineering problems we face every day.


---related links---

http://twitter.com/peckjon
http://linkedin.com/in/peckjon
https://github.com/peckjon