PromptTuner

PromptTuner MCP is a server that analyzes, refines, and optimizes prompts to improve AI understanding and responses. It supports multiple LLM providers, offers practical tooling to fix prompts, and can run in HTTP mode or as a local stdio process for integration with MCP clients.

How to use

You can run PromptTuner MCP locally and connect to it from an MCP client or workflow. Use the HTTP endpoint for testing or quick integration, or run a local stdio process if your MCP client supports a direct command. You’ll benefit from improved prompt quality, faster refinements, and multi-provider support.

Using a local stdio configuration lets your client launch the MCP server as a child process. The example below shows how to start the server using Node and the built dist file, along with required environment variables for your chosen LLM provider.

How to install

Prerequisites: ensure you have Node.js installed on your system. You can verify by running node -v and npm -v. If you plan to run via Docker, have Docker and Docker Compose available.

git clone https://github.com/j0hanz/prompttuner-mcp.git
cd prompttuner-mcp
npm install
npm run build

Additional configuration and usage notes

Configure your MCP client to run PromptTuner MCP either as a local stdio process or by pointing to the HTTP endpoint. The following configurations illustrate typical setups you can adapt.

// Claude Desktop configuration (stdio)
{
  "mcpServers": {
    "prompttuner": {
      "command": "node",
      "args": ["/path/to/prompttuner-mcp/dist/index.js"],
      "env": {
        "LLM_PROVIDER": "openai",
        "OPENAI_API_KEY": "sk-..."
      }
    }
  }
}

HTTP mode for testing or integration with HTTP-based clients exposes the MCP endpoint at a local address. Use a POST request to the following URL to send prompts to refine or analyze them.

# Start in HTTP mode (experimental)
npm run start:http
# Server runs at http://127.0.0.1:3000/mcp

Examples and quick start

Here is a typical sequence to get up and running quickly: clone the project, install dependencies, build, and start in the desired mode.

git clone https://github.com/j0hanz/prompttuner-mcp.git
cd prompttuner-mcp
npm install
npm run build

Available tools

refine_prompt

Fix grammar, improve clarity, and apply optimization techniques to a given prompt. Includes intelligent caching to speed up repeated refinements.

analyze_prompt

Score prompt quality across dimensions and provide actionable improvement suggestions.

optimize_prompt

Apply multiple techniques sequentially for comprehensive prompt improvement and return before/after scores and diff.

detect_format

Identify target AI format with confidence score and provide recommendations.

compare_prompts

Compare two prompts side-by-side with scores, diffs, and recommendations.

validate_prompt

Pre-flight validation to check issues, estimate tokens, and detect anti-patterns and security risks before use.