Langfuse Prompt Management

Exposes Langfuse's centralized prompt repository through the Model Context Protocol's Prompts specification, implementing the prompts/list endpoint with pagination support. The server translates Langfuse's REST API responses into MCP's JSON-RPC message format, filtering prompts by production label and returning metadata (name, description, version) for client-side discovery. Uses stdio transport with JSON-RPC 2.0 for bidirectional communication with MCP clients like Claude Desktop and Cursor IDE.

Retrieves a specific prompt from Langfuse by name and compiles it with user-provided variables, handling both text and chat prompt types. The server extracts template variables from Langfuse's prompt structure (using pattern matching or AST-like parsing), validates that all required variables are provided, and returns a fully compiled prompt ready for LLM inference. Supports Langfuse's native prompt types (text prompts and chat message arrays) and transforms them into MCP's standardized message format for consumption by MCP clients.

Provides two complementary interfaces to the same underlying Langfuse prompt repository: the MCP Prompts specification (primary, standards-based) and MCP Tools (compatibility fallback). The server implements both prompts/list and prompts/get endpoints alongside get-prompts and get-prompt tools, allowing clients with different MCP capability support to access the same prompt data. This dual interface pattern is handled at the request routing layer, where incoming JSON-RPC requests are dispatched to the appropriate handler based on the method name.

Automatically filters Langfuse prompts to expose only those labeled as 'production', preventing clients from accidentally using draft, experimental, or outdated prompt versions. This filtering is applied at the listing and retrieval layers — the prompts/list endpoint only returns production-labeled prompts, and prompts/get will reject requests for non-production prompts. The filtering logic is implemented in the request handlers and uses Langfuse's native label metadata to determine eligibility, ensuring that only vetted, released prompts are accessible through the MCP interface.

Implements the Model Context Protocol's stdio transport layer, communicating with MCP clients via standard input/output using JSON-RPC 2.0 message format. The server runs as a Node.js process that reads JSON-RPC requests from stdin, processes them through the appropriate handler (prompts/list, prompts/get, or tools), and writes JSON-RPC responses to stdout. This transport mechanism is language-agnostic and allows the MCP server to be spawned by any client that supports stdio-based process communication, including Claude Desktop, Cursor IDE, and custom MCP consumers.

Manages authentication to the Langfuse API using environment variables (LANGFUSE_SECRET_KEY and LANGFUSE_PUBLIC_KEY) and constructs authenticated HTTP requests to Langfuse's REST endpoints. The server reads credentials from the environment at startup, validates their presence, and includes them in all outbound API calls to Langfuse. This credential management is centralized in the server initialization, eliminating the need for clients to handle Langfuse authentication directly and allowing the MCP server to act as a trusted intermediary between MCP clients and Langfuse.

Handles two distinct Langfuse prompt types (text prompts and chat prompts) and transforms them into MCP's standardized message format. Text prompts are returned as plain strings, while chat prompts are parsed as arrays of messages with roles (system, user, assistant) and compiled with variable substitution. The server detects the prompt type from Langfuse's metadata and applies the appropriate transformation logic, ensuring that both prompt types are accessible through the same MCP interface. Chat prompts are particularly important for multi-turn conversations and role-based message construction in LLM applications.

Integrates with Langfuse's REST API by constructing HTTP requests to Langfuse endpoints (typically /api/prompt endpoints for listing and retrieving prompts). The server uses a configurable base URL (defaulting to Langfuse's hosted API but supporting self-hosted instances) and constructs authenticated requests with proper headers and query parameters. This integration layer abstracts away the details of Langfuse's API structure, allowing the MCP server to act as a transparent proxy that translates MCP requests into Langfuse API calls and transforms responses back into MCP format.