Lints AI prompts like a code quality tool. Exposes a single MCP resource that analyzes system prompts, agent instructions, and tool descriptions for contradictions (like "be concise" paired with "be thorough"), redundant sentences, ambiguous pronouns, oversized code examples, and politeness fluff that wastes tokens. Returns a scored report with line numbers and actionable suggestions. Also includes a CLI mode with JSON output and a trim flag that strips fluff while preserving fenced code blocks and quoted strings. Useful when you're iterating on complex system prompts and want to catch silent breaking issues before they burn tokens on every request. Runs purely static analysis, no LLM calls.
claude mcp add --transport stdio sean-sunagaku-promptlint-mcp uvx promptlint-mcp