clarifyprompt-mcp

Purpose

ClarifyPrompt MCP is a context-aware MCP prompt compiler. It receives raw, vague prompts, gathers relevant context, parses the user's actual intent, and then returns a prompt version optimized specifically for the target platform, covering 61 platforms including Midjourney, GPT Image 2, Veo, Runway, Kling, Seedance, Higgsfield, ElevenLabs, Suno, Claude, ChatGPT, Cursor, and more.

Main Capabilities

• Platform adaptation: Generates prompts for different AI platforms that conform to their native syntax, parameters, structure, and fundamentals. For example, Midjourney uses ", --ar/--v/--s/--q " flags, while DALL-E and Nano Banana use natural-language descriptions without flags.
• Workspace context awareness: Extracts information from workspace signals (such as CLAUDE.md, AGENTS.md, .cursorrules, package.json, git status, current files), parses the user's true intent, and then generates the prompt.
• Composable pipeline: Supports the complete ", clarify → ground/optimize → critique → revise " process, and can be completed in a single call via ", compose_prompt ".
• Critique and revision: ",critique_prompt " provides six default evaluation dimensions: clarity, specificity, intent_alignment, format_fitness, length_appropriateness, plain_language, and supports an automatic revision loop.
• Platform fact validation: Built-in platform packs are continuously updated, and each entry carries a ", verifiedAt " date to ensure prompts are based on the latest vendor documentation.
• MCP Apps interactive panel: In hosts that support the ", io.modelcontextprotocol/ui " extension (such as Claude Desktop, ChatGPT, Cursor, VS Code), ",compose_prompt " can render a real-time panel showing a comparison of the original and optimized prompts, six-dimensional scores, pipeline stages, and provide Accept / Revise actions.
• Latest model compatibility: The client can correctly send parameters, handle cases where models such as claude-sonnet-5, gpt-5/o-series, and Gemini reject temperature or max_tokens, and can learn from 400 errors and retry.
• Portable text output by default: For chat, document, and code prompts, if no platform is specified, output a platform-neutral format instead of defaulting to Claude's XML tag style.
• Plain-language rewriting: Optimized prompts use everyday wording, avoid formal vocabulary drift, and derive specificity from concrete details rather than fancy synonyms.

How to Use

Users send a raw prompt to ClarifyPrompt, which calls the ", optimize_prompt " tool, specifies the category and platform, and returns the optimized prompt. For interactive use, ",compose_prompt " can call the full pipeline and render the panel. Users need to configure an OpenAI-compatible API endpoint (such as ", LLM_API_URL"), optimized by default for local small models. For specific installation and configuration steps, refer to the GitHub repository's README.

Suitable Scenarios

• Need to generate prompts for multiple AI platforms that conform to their respective dialects.
• In a coding environment, automatically generate context-relevant prompts based on the current project files and dependencies.
• Want prompts that reflect the platform's latest features (such as Midjourney V8.2, Runway Gen-4.5, etc.) to avoid outdated model knowledge.
• Scenarios that require systematic evaluation and iteration of prompt quality.

Limitations

• The prompt compiler itself does not generate images, videos, or other content; it only compiles prompts, and users need to send the result to the target model.
• Output depends on the underlying model; wording generated by different models may differ, but the structure remains consistent.
• Built-in fact packs are continuously updated but may still lag behind the latest platform changes; the ",verifiedAt " date can be used for tracking.
• Some platforms (such as Sora) have been discontinued, and related support has been removed.
• You need to configure an OpenAI-compatible API endpoint yourself, and reasoning models may require increasing LLM_TIMEOUT_MS Timeout setting.