Rizel Scarlett from GitHub explains practical techniques for responsible AI coding with GitHub Copilot and MCP, focusing on safer prompt engineering and developer empowerment.

Responsible Vibe Coding with GitHub Copilot and Model Context Protocol

Speaker

• Rizel Scarlett, Tech Lead of Open Source Developer Relations, Block

Overview

This session at GitHub Universe 2025 dives into the concept of ‘vibe coding’—using AI agents like GitHub Copilot to write code in a freeform, exploratory way. Rizel addresses the risks developers face, such as agents overwriting files, leaking API keys, or hallucinating broken processes, and introduces the Model Context Protocol (MCP) as a solution for more structured, context-aware AI development.

Key Topics

• Vibe Coding vs Vibe Engineering: Defining the difference between letting AI take the lead and guiding its output responsibly.
• Three Principles for Responsible AI Coding:
  1. Power: Selecting appropriate LLM capabilities for each use case.
  2. Context: Supplying precise, relevant information to AI agents to mitigate errors and unintended actions.
  3. Control: Remaining actively engaged as the human ‘pilot,’ using protocols like Agent Client Protocol to assert oversight.
• Model Context Protocol (MCP):
  • Adds structure and contextual awareness to prompt engineering.
  • Enables safer, more observable interactions with AI agents.
• GitHub MCP Server:
  • Facilitates context and control for developers during AI coding.
  • Demoed tools:
    • Goose (creating issues via MCP)
    • MCP UI (visualizes team workload)
• Agent Client Protocol:
  • Allows for ongoing monitoring and engagement with AI agents to prevent automated mistakes.

Demos

• Creating GitHub issues with Goose through MCP
• Visualizing team workloads using MCP UI

Conclusion

• Emphasizes that developers remain the pilot, not just the passenger, when working with AI agents.
• Encourages leveraging the new protocols to keep productivity high and errors low.

References

• GitHub Universe 2025 session list: Playlist
• Blog and updates: GitHub Blog
• Agent Client Protocol and MCP documentation (if available via GitHub)

Links

• GitHub Copilot: https://github.com/features/copilot
• MCP overview and resources: (Search official MCP documentation on GitHub)

This content is tailored for developers interested in practical, responsible adoption of AI coding assistants and new protocols for safer, more effective workflow.