Rubber Duck Thursdays - Building a turn-based-game MCP server

Key Concepts

• Model Context Protocol (MCP): A specification for enabling AI agents to interact with applications through structured tool calls and data exchange.
• Elicitation: A process within MCP where the server requests additional information from the user to complete a tool call.
• Tool Calls: Functions or actions that the MCP server can execute, triggered by user input and interpreted by the AI agent.
• GitHub Copilot Coding Agent: An AI assistant that can automate code refactoring, bug fixing, and other development tasks.
• Dev Container: A configuration file that defines a consistent development environment, ensuring repeatable builds and reducing "works on my machine" issues.
• Dependabot: A GitHub tool that automatically creates pull requests to update dependencies, improving security and staying up-to-date with the latest versions.
• Copilot Instructions File: A file that provides context and guidelines to GitHub Copilot, improving its understanding of the project and code style.

Main Topics and Key Points

1. Introduction and Updates

• The session is a "Rubber Duck Thursday" (RDT) focused on building an MCP server for a turn-based game.
• Recent GitHub updates are reviewed, including:
  • npm Trusted Publishing with OIDC: Enables publishing to npm using OpenID Connect, eliminating the need for long-lived tokens and improving supply chain security.
  • Expanded Repository Creation Experience: An improved UI for creating repositories with easier access to settings.
  • Copilot Chat Repository Management Skills: Copilot Chat can now create, update, and push files, create branches, and merge pull requests.
  • GitHub Actions API: New REST APIs for managing settings like approval for running fork pull requests and setting artifact retention periods.
  • Claude Opus 4.1: The latest version of the Claude Opus model is now available in public preview for GitHub Copilot Enterprise and Pro Plus plans.
  • Secret Protection Updates: Expanded default pattern support and additional validators for secret scanning.
  • Copilot Coding Agents Improvements: Copilot coding agents now only respond when explicitly mentioned with @copilot.
  • Spark Improvements: Enhanced reliability, seeds, data, and performance updates for Spark.
• The presenter encourages viewers to check out the full change log at gh.io/rdt/changelog.

2. Love of Code Hackathon and AI Skills Challenge

• The "Love of Code" hackathon is highlighted, encouraging viewers to participate in joyful, ridiculous, and wildly creative projects. The hackathon runs from July 16th to September 22nd and includes categories like hardware, AI, terminal talent, games, web apps, and "everything but the kitchen sink." The URL is gh.io/love-of-code.
• An AI skills challenge is presented, where participants can use AI to explore repositories. The challenge is open until August 21st, and participants can win a GitHub shop voucher. The URL is gh.io/rdt-skills-1.

3. Turn-Based Game MCP Server Development

• The main focus is on continuing the development of a turn-based game MCP server, specifically for tic-tac-toe and rock-paper-scissors.
• The presenter shares the GitHub repository for the project: gh.io/rdt/game-mcp.
• A demonstration of the current state of the MCP server is given, showcasing how it interacts with GitHub Copilot to play games.
• The concept of elicitation is introduced, where the MCP server requests additional information from the user to complete a tool call.
• The presenter demonstrates how elicitation is used to gather information about the game, such as the difficulty level, who goes first, and the player's name.

4. Elicitation Implementation and Technical Debt

• The presenter explains the implementation of elicitation in the MCP server, including the use of schemas to define the required inputs.
• Technical debt is acknowledged, specifically the duplication of tool definitions and handlers.
• The presenter outlines a plan to consolidate the tool definitions and handlers, making the code more maintainable.
• The presenter uses GitHub Copilot to generate an issue and begin implementing the changes.

5. Repository Setup and Best Practices

• The presenter discusses the importance of setting up a repository with certain features from the beginning, including:
  • Dev Container: Ensures a repeatable development environment.
  • Dependabot: Keeps dependencies up-to-date.
  • Copilot Instructions File: Provides context and guidelines to GitHub Copilot.
  • CI/CD Workflow: Automates builds, tests, and linting.
• The presenter also recommends turning on certain repository settings, such as:
  • Always suggest updating pull request branches.
  • Allow automerge.
  • Automatically delete head branches.
• The presenter emphasizes the importance of a code of conduct, contributing file, security guidelines, and a well-written README.

6. Tool Consolidation and Refactoring

• The presenter uses GitHub Copilot to consolidate the tool definitions and handlers, reducing duplication and improving maintainability.
• The presenter demonstrates how the consolidated tool calls are used in the MCP server.
• The presenter identifies additional areas for refactoring, such as duplicated constants and difficulty styling.

7. Copilot Coding Agent and Codebase Cleanup

• The presenter uses GitHub Copilot Coding Agent to automate the codebase cleanup, specifically to address the duplicated constants and difficulty styling.
• The presenter reviews the plan generated by the coding agent and provides feedback.
• The presenter monitors the progress of the coding agent as it refactors the code.

8. Bugs and Future Work

• The presenter identifies a bug where elicitation is always executed, even if the information is provided in the initial message.
• The presenter acknowledges that there are other bugs that need to be fixed, such as the infinite loop in the rock-paper-scissors game.
• The presenter outlines the plan for future work, which includes fixing the bugs and continuing the codebase cleanup.

9. Conclusion and Announcements

• The presenter summarizes the work that was done during the session, including the elicitation implementation, tool consolidation, and codebase cleanup.
• The presenter announces that there will be a break from the stream for the next two weeks.
• The presenter encourages viewers to participate in the "Love of Code" hackathon.
• The presenter announces that Cadia will be streaming later today, focusing on the GitHub MCP server.

Notable Quotes

• "Elicitation is basically a fancy word a fancy way of saying um if we don't have all the information we need to go and get it."
• "I'm not uh I'm not I don't even know what the right right word is. So, there we go. I definitely don't know any everything."
• "The read me as well, that is like the shop window. that is the kind of the the front of your repository, right? And it gives you all the information that you kind of need."

Technical Terms and Concepts

• MCP Server: A server that implements the Model Context Protocol, allowing AI agents to interact with applications.
• MCP Client: A client that communicates with the MCP server, typically running within an IDE or other application.
• Tool Definition: A description of a tool that the MCP server can execute, including its name, description, and inputs.
• Tool Handler: A function that is executed when a tool is called.
• Schema: A definition of the structure and data types of the inputs and outputs of a tool.
• GitHub Actions: A CI/CD platform that automates builds, tests, and deployments.
• Linting: A process of analyzing code for potential errors and style violations.
• Minimax Algorithm: A decision-making algorithm used in game theory to find the optimal move for a player.

Logical Connections

• The session begins with a review of recent GitHub updates, providing context for the ongoing development of the MCP server.
• The presenter then introduces the "Love of Code" hackathon and AI skills challenge, encouraging viewers to get involved in the community.
• The main focus of the session is on the development of the turn-based game MCP server, specifically the implementation of elicitation and the consolidation of tool calls.
• The presenter then discusses the importance of setting up a repository with certain features from the beginning, providing best practices for software development.
• The presenter uses GitHub Copilot Coding Agent to automate the codebase cleanup, demonstrating the power of AI in software development.
• The session concludes with a summary of the work that was done and announcements about future streams and events.

Data, Research Findings, or Statistics

• The session mentions that Claude Opus 4.1 is available to Copilot Enterprise and Pro Plus plans.
• The session mentions that the GPU machine type in code spaces will be deprecated on August 29th.

Section Headings

1. Key Concepts
2. Main Topics and Key Points
3. Notable Quotes
4. Technical Terms and Concepts
5. Logical Connections
6. Data, Research Findings, or Statistics

Synthesis/Conclusion

The session provides a detailed overview of the development of a turn-based game MCP server, highlighting the implementation of elicitation, the consolidation of tool calls, and the use of GitHub Copilot Coding Agent for codebase cleanup. The presenter also shares valuable insights into repository setup and best practices, encouraging viewers to get involved in the community and participate in the "Love of Code" hackathon. The session demonstrates the power of AI in software development and provides actionable insights for developers looking to build their own MCP servers.