Hooks into Claude CLI to automatically score every prompt you write against your actual codebase. Runs evaluate_prompt behind the scenes, scanning your file tree, git state, and TypeScript symbols to grade specificity, context completeness, task clarity, and file anchoring on a 100-point scale. Returns concrete suggestions and an AI-rewritten version of your prompt that references real paths and verified function names from your project. Also exposes get_context for project summaries and prompyai_toggle to pause scoring. Works immediately with npx, no config needed. If you set ANTHROPIC_API_KEY it uses Haiku directly for rewrites, otherwise it feeds context back to Claude itself. Useful when you're working in a large codebase and want your prompts grounded in what actually exists rather than what you think exists.
claude mcp add --transport stdio samouh-waleed-prompyai uvx prompyai