playdate-mcp is a Model Context Protocol server that gives AI agents direct access to Playdate development workflows. It exposes build, simulator, test, and metadata operations in a form that coding agents can call safely.
The server exposes tools for compiling a Playdate project, launching the simulator, running desktop Lua tests, surfacing structured build errors, and reading pdxinfo metadata. The result is a tighter feedback loop for Playdate-focused coding sessions.
There was no equivalent tool in the Playdate ecosystem for AI-native development workflows. I wanted a way for agents to build games, run the simulator, execute Lua tests, and inspect project files without treating everything as raw shell automation.
The project is a small TypeScript MCP server built on @modelcontextprotocol/sdk with Zod schemas for input validation. Each tool resolves paths, executes a well-scoped Playdate workflow, and returns structured text content that an agent can reason about.
I kept the initial tool surface narrow and practical: build, run, test, and inspect. That keeps the server immediately useful while preserving room to expand once the core workflows prove stable.
The hardest part is translating Playdate’s local toolchain expectations into predictable agent-facing behavior. Strong path validation and structured error responses help keep failures understandable.
playdate-mcp establishes a foundation for AI-assisted Playdate development and fills a gap that previously required ad hoc terminal scripting.
