MCP Prompt Server

MCP Prompt Server is an MCP server that provides preconfigured prompt templates exposed directly as callable tools for AI-assisted editors such as Cursor, Windsurf, and Cline. Instead of serving standard MCP prompts, it converts custom and default prompt files into executable tools, allowing agents to execute structured tasks with parameterized inputs. Developers and software engineers use this server to standardize development routines, including code reviews, documentation creation, refactoring recommendations, test generation, and architectural design evaluations. Built on Node.js using standard input and output (stdio) transport, the server parses template definitions formatted in YAML or JSON files located in a local prompts directory. Each prompt supports dynamic argument placeholders, giving coding agents the exact contextual instructions required to process source snippets or user requirements. In addition to serving templates, the server includes administrative utilities to query active templates and dynamically reload modifications from disk without restarting the host editor process.

Category: AI & LLM Tooling

Tags: productivity, prompts, templates, workflow

Visit MCP Prompt Server

How to install and configure MCP Prompt Server

1. Clone the repository and install dependencies: bash cd prompt-server npm install 2. To verify the server locally, run: bash npm start 3. Register the server in Cursor by adding it to your Cursor MCP configuration file: json { "servers": [ { "name": "Prompt Server", "command": ["node", "/path/to/prompt-server/src/index.js"], "transport": "stdio", "initialization_options": {} } ] } 4. Alternatively, register in Windsurf by editing ~/.codeium/windsurf/mcp_config.json: json { "mcpServers": { "prompt-server": { "command": "node", "args": [ "/path/to/prompt-server/src/index.js" ], "transport": "stdio" } } } Replace /path/to/prompt-server with your actual directory path, then restart your editor or refresh the MCP tools panel.

What you can do with MCP Prompt Server

• Running structured code reviews across various languages by passing code blocks directly into the code_review template tool. - Generating Markdown API reference documentation from source code functions and class definitions using the api_documentation tool. - Executing standardized code refactoring workflows by sending target snippets and modernization goals to the code_refactoring prompt. - Creating automated test cases for existing implementation logic via the test_case_generator prompt template tool. - Adding custom team prompts in YAML format to standardize software architecture analysis across development team members.

Key facts

• https://github.com/gdli6177/mcp-prompt-server
• AI & LLM Tooling
• productivity, prompts, templates, workflow

Part of MCP Servers

What is MCP Prompt Server?

MCP Prompt Server is a Model Context Protocol server that wraps predefined prompt templates as MCP tools rather than standard prompts. It is designed to help AI coding clients like Cursor, Windsurf, and Cline invoke structured development workflows, such as code reviews and documentation generation, using dynamic input arguments.

Which MCP clients work with MCP Prompt Server?

The server works with any MCP-compatible environment supporting stdio communication. It is specifically optimized and documented for integration with code editors and development environments such as Cursor, Windsurf, and Cline.

How do I add custom prompts to MCP Prompt Server?

You can add new prompt templates by creating YAML or JSON files inside the src/prompts directory. Each file requires a name, description, optional arguments list, and message definitions with placeholder tokens. You can then invoke the reload_prompts tool to make the new template immediately available.

What tools are built into MCP Prompt Server?

The server provides administrative tools named reload_prompts to refresh templates from the filesystem and get_prompt_names to retrieve all registered prompt names. It also exposes individual tools generated from template files, including code_review, api_documentation, code_refactoring, test_case_generator, and project_architecture.