MLflow Prompt Registry MCP Server

What is MLflow Prompt Registry MCP Server?

It is a Model Context Protocol (MCP) server that connects to MLflow Prompt Registry, allowing AI assistants (like Claude Desktop) to discover and load prompt templates stored in MLflow. It implements the MCP Prompts specification for retrieving and using reusable prompt templates.

How to use MLflow Prompt Registry MCP Server?

Install MLflow (≥2.21.1), start an MLflow server, create a prompt template in MLflow, then build this MCP server (npm install && npm run build). Finally, add it to Claude Desktop by editing claude_desktop_config.json with the node command pointing to dist/index.js and setting the MLFLOW_TRACKING_URI environment variable to your MLflow server address.

Key features of MLflow Prompt Registry MCP Server

• Exposes two tools: list-prompts and get-prompt
• Lists available prompt templates with optional cursor and filter
• Retrieves and compiles a specific prompt by name with variable arguments
• Enables Claude Desktop to discover MLflow prompt templates
• Follows the MCP Prompts specification

Use cases of MLflow Prompt Registry MCP Server

• Instructing Claude for repetitive coding or documentation tasks
• Using shared, version-controlled prompt templates from MLflow
• Automating common workflows by loading prompts on demand
• Integrating MLflow prompt management with MCP-compatible clients

FAQ from MLflow Prompt Registry MCP Server

What are the runtime dependencies?

Node.js and npm to build and run the server, plus an MLflow server (≥2.21.1) hosting the Prompt Registry.

Where do prompt templates live?

Prompt templates are stored and managed in the MLflow server (Prompt Registry). The MCP server reads them via the MLFLOW_TRACKING_URI environment variable.

How is the server configured for transport?

The server connects over stdio (standard input/output). No HTTP or WebSocket transport is mentioned.

Does the server support authentication?

The README does not mention any authentication. It assumes a local or accessible MLflow server without auth.

Are there any known limits?

The README does not list limits. Output size depends on prompt template size and number of prompts returned by MLflow.