AgentForge

Turns vague coding requests into structured, agent-ready prompts through a single MCP tool. You call agentforge_refine_prompt with a rough task description, pick your target agent (Claude Code, Cursor, Aider, etc.), and get back a prompt that's been scored across 12 quality dimensions. The engine extracts requirements, formats for your specific tool's expectations, and checks everything before returning. Runs through npx with an API key from agentforge.sciscale.org. Free tier gives you three refinements per day; Pro unlocks unlimited use and auto-refines until the prompt scores above 90. The whole engine lives server-side, so no logic ships in the package. Useful when you're tired of the clarification loop with your coding agent and want to hand it something it can execute in one pass.