Adding MCP Apps Support to Apollo MCP Server with Agentic Coding and Goose

Adding MCP Apps Support to Apollo MCP Server with Agentic Coding and Goose

Amanda Martin, a developer advocate, successfully added experimental support for MCP Apps to the Apollo MCP Server, a GraphQL-to-MCP tool, despite having no prior experience writing Rust – the language the server is built in. This was achieved using an agentic development workflow powered by the Goose CLI and Desktop application.

The project demonstrates the potential of AI-assisted coding to lower the barrier to entry for contributing to complex codebases. Traditionally, contributing to a project requires proficiency in its core languages, but this example showcases a path to contribution even without that expertise, potentially accelerating development cycles and fostering wider community involvement.

Why This Matters

Current software development often relies on specialized expertise, creating bottlenecks and increasing costs. While agentic coding promises to mitigate this, real-world integration presents challenges. This project highlights the importance of iterative planning and testing when using AI agents, as unvalidated changes can introduce subtle bugs that are difficult to identify, potentially causing downtime or data inconsistencies – issues that can cost organizations significant resources.

Key Insights

• Goose CLI & Desktop: Used to facilitate agentic coding and debugging, enabling contributions to a Rust codebase without direct Rust coding experience.
• Iterative Planning: Breaking down the task into smaller, testable chunks significantly improved the agent’s success rate and reduced debugging time.
• MCP Apps Draft Spec: The project focused on adding support for an evolving standard, highlighting the need for adaptability in agentic workflows as specifications change.

Practical Applications

• Internal Tooling: Apollo can now leverage MCP Apps to create internal dashboards like the Luma Community Analytics tool for tracking event metrics.
• Pitfall: Relying solely on agent-generated code without thorough testing can lead to subtle bugs, as demonstrated by the discovered query parameter incompatibility.

References:

• https://dev.to/amandamartindev/how-i-added-experimental-mcp-apps-support-to-apollo-mcp-server-with-goose-2hjl