mcp-prompt

MCP Prompt Server exposes a collection of pre-defined prompt templates as MCP tools. These tools are designed to be consumed by editors like Cursor and Windsurf, enabling you to run common tasks such as code review, API documentation generation, and code refactoring directly through the MCP interface. The server loads prompts from the src/prompts directory and presents each prompt as a tool that can be invoked with a structured payload. You can also reload prompts at runtime using the provided management tools, ensuring your tooling stays up-to-date as you add or modify templates. The server’s design emphasizes dynamic parameter replacement, so you can tailor prompts with specific language, code, or formatting requirements before sending them to your editor or agent.

Available prompt tools (as defined in the repository) include code_review, api_documentation, code_refactoring, test_case_generator, and project_architecture. Each tool accepts an arguments object that encodes the necessary inputs for the task (e.g., language, code snippet, or project context) and returns a structured prompt payload ready to be consumed by your MCP-enabled editor or agent. In addition to prompts, the server offers management actions like reload_prompts and get_prompt_names to manage and inspect the available tools. For integration, configure the MCP client in your editor to point to the node-based server (e.g., via a standard stdio transport), and reference the appropriate command and path to the server entry point.

Prerequisites:

• Node.js (v14+ recommended) and npm/yarn installed
• Access to the repository containing the prompt-server project

Installation steps:

1. Clone the repository

git clone https://github.com/your-org/path-to-repo.git
cd path-to-repo

2. Install dependencies

cd prompt-server
npm install

3. Run the server

npm start

4. Connect clients (Cursor/Windsurf) to the MCP server using the appropriate configuration, typically pointing to the node entry at src/index.js as shown in the README examples.

Notes:

• Ensure that the src/prompts directory contains the YAML/JSON prompt definitions you intend to expose as tools.
• To add new prompts, drop a YAML/JSON file into src/prompts and restart or reload the prompts via the management tool (reload_prompts) if provided by the server.
• If you modify the server code, restart the server to apply changes.

Tips and considerations:

• The server exposes prompts as MCP tools rather than MCP prompts themselves; use the tool API (e.g., get_prompt_names) to enumerate available prompts.
• When integrating with editors, ensure the transport (stdio) is correctly configured in the editor's MCP configuration file.
• If you're deploying in a container or CI environment, consider pinning Node.js versions and scripts to ensure reproducible startup.
• Watch for environment-specific paths (e.g., /path/to/prompt-server) and replace with your actual deployment path.
• If prompts rely on dynamic parameters, ensure your editor or agent supplies those parameters in the arguments section to avoid runtime errors.