MCP Server for Langfuse Prompts

The Langfuse Model Context Protocol (MCP) Server enables AI assistants and agents to interact with your Prompt Management system. The server is built directly into Langfuse at /api/public/mcp (streamableHttp), no external setup or build steps required.

Available Tools

The Langfuse MCP Server provides five tools for comprehensive prompt management in Claude Code and other MCP clients:

Features

Read Operations

• getPrompt - Fetch a specific prompt by name

  • Optional label parameter (e.g., “production”, “staging”) - defaults to “production”
  • Optional version parameter to get a specific version number
  • Returns compiled prompt with metadata and configuration

• listPrompts - Browse all prompts in your project

  • Optional filtering by name, tag, or label
  • Cursor-based pagination for large prompt libraries
  • Returns prompt metadata including available versions and labels

Write Operations

• createTextPrompt - Create a new text prompt version

  • Simple string content with {{variable}} template syntax
  • Optional labels (e.g., ["production", "staging"])
  • Optional config object for model settings
  • Optional tags for organization
  • Optional commit message to document changes
  • Automatically increments version number

• createChatPrompt - Create a new chat prompt version

  • OpenAI-style message format with role and content
  • Supports system, user, and assistant message roles
  • Template variables work in message content
  • Same label, config, tag, and commit message options as text prompts

• updatePromptLabels - Manage labels across prompt versions

  • Move labels between versions (e.g., promote staging to production)
  • Labels are unique—setting a label on one version removes it from others
  • Cannot modify the auto-managed latest label
  • Useful for deployment workflows and version control

Setup

See the MCP Server setup guide for complete instructions on authentication and client configuration.

Example Workflows

Create a New Prompt

Ask your AI agent:

“Create a new text prompt called ‘customer-email’ with a friendly greeting template that includes name and product variables. Tag it as ‘draft’ and add a production label.”

The agent will use createTextPrompt to create the prompt version.

Promote a Prompt to Production

“Move the production label from version 2 to version 3 of the customer-email prompt”

The agent will use updatePromptLabels to update the label assignment.

Iterate on a Chat Prompt

“Create a new version of the code-review prompt with improved system instructions and add an example assistant message showing the review format”

The agent will use createChatPrompt to create a new version with enhanced content after having fetched the prompt with getPrompt.

Feedback

Share your experience with the Langfuse MCP server in our GitHub Discussion. We’d love to hear your feedback and use cases.

Learn More

• MCP Server Overview - Complete MCP server documentation
• Prompt Management - Learn about Langfuse prompt features
• Model Context Protocol - Official MCP documentation