Home / MCP / Claude Prompts MCP Server
Provides hot-reloadable prompts and structured reasoning for building programmable AI assistants with chain workflows.
Configuration
View docs{
"mcpServers": {
"claude_prompts": {
"command": "npx",
"args": [
"-y",
"claude-prompts-server"
]
}
}
}You have a ready-to-use MCP server that turns your prompt library into a version-controlled, hot-reloadable engine. It renders structured prompts with frameworks, gates, and chain workflows so you can build and test AI assistant reasoning quickly without restarting after every edit.
You connect to the Claude Prompts MCP Server from your MCP client to run and test prompt pipelines. Start a local server, point your client at it, and then compose workflows using fast, observable steps. Use the provided examples to run prompts, chain steps, and apply quality gates so Claude self-checks as it reasons.
Prerequisites: you need Node.js and npm installed on your system. You can also use npx to run the server without a local clone.
Option A — clone and build (recommended for prompt management)
git clone https://github.com/minipuft/claude-prompts-mcp.git
cd claude-prompts-mcp/server
npm install
npm run build
npm run start:stdioOption B — run directly from npm without cloning
# Run directly without global installation
npx claude-prompts-server
# Or install globally for easy access
npm install -g claude-prompts-server
claude-prompts-server --helpNote: When installed via npm, prompt files may reside under your node_modules directory. The server can load external prompt configurations (via MCP_SERVER_ROOT or by placing config.json in your current working directory). Direct editing of bundled prompts is not recommended for npm installations.
Connect Claude Desktop by adding an MCP server entry that runs the CLI runtime for this server. The examples below show how to use npx (no clone) and a source build on Windows or macOS/Linux.
{
"mcpServers": {
"claude-prompts": {
"command": "npx",
"args": ["-y", "claude-prompts-server"]
}
}
}If you build from source, run the local Node.js runtime pointing to the built index. On Windows, use the dist/index.js path; on macOS/Linux, use the corresponding dist/index.js path for your environment.
{
"mcpServers": {
"claude-prompts": {
"command": "node",
"args": ["C:\\path\\to\\claude-prompts-mcp\\server\\dist\\index.js"]
}
}
}Restart Claude Desktop after making changes. In the input bar, type a command like prompt_manager list to verify the server is reachable and the available prompts can be listed.
Key concepts include hot reload of prompts, a chain workflow syntax, and the use of frameworks and gates to guide and validate Claude’s reasoning. You can manage prompts with prompt_manager, run templates with prompt_engine, and adjust guidance through framework modifiers.
Typical workflow: edit a prompt, describe the issue, have Claude apply a fix via prompt_manager, then test immediately with prompt_engine. This creates a fast feedback loop to improve prompts without restarting servers.
- If the server doesn’t respond, verify Node.js and npm versions are compatible with the project. Re-run npm install after pulling updates.
- Use the provided examples to validate CLI integration with Claude Desktop, ensuring the MCP server entry is correctly configured and the command/args array matches what you run locally.
Prompts are treated as version-controlled assets. When you modify prompts, you gain an auditable history. Use gated checks to ensure outputs meet quality and safety criteria before producing final results.
Manages prompt lifecycle by updating, editing, and organizing prompts as versioned assets within the MCP workflow.
Executes prompt templates and coordinates chain steps, enabling iterative runs and testing of prompt responses.