@Jurislm/Langfuse

MCP (Model Context Protocol) server for Langfuse — provides 50 tools for prompt management, observability, instance/organization management, and metrics tracking via natural language.

Tools (50)

Prompt Management (6 tools)

• listPrompts — List and filter prompts with pagination, labels, tags
• getPrompt — Fetch a specific prompt version with resolved dependencies
• createTextPrompt — Create a text prompt version (supports {{variable}} syntax)
• createChatPrompt — Create a chat prompt with role-based messages
• getPromptUnresolved — Fetch a prompt without resolving dependencies (for debugging composition)
• updatePromptLabels — Update labels for a prompt version (labels are unique across versions)

Traces (2 tools)

• listTraces — List traces with pagination and optional filters (name, userId, tags, timestamps)
• getTrace — Get a single trace with full details, observations, and scores

Observations (2 tools)

• listObservations — List observations (generations, spans, events) with filters by type and name
• getObservation — Get a single observation with complete details (input, output, model, usage, duration)

Scores (2 tools)

• createScore — Create a score for a trace or observation (for evals, feedback, quality metrics)
• listScores — List scores with pagination and optional filters

Sessions (1 tool)

• listSessions — List sessions that group related traces together across time ranges

Datasets (8 tools)

• listDatasets — List datasets for evals and benchmarking with search and pagination
• getDataset — Get dataset details with metadata and item counts
• createDataset — Create a new dataset for evals or benchmarking
• listDatasetItems — List items in a dataset with pagination (input/output pairs)
• createDatasetItem — Add an input/output pair to a dataset for eval testing
• getDatasetItem — Get a single dataset item by ID
• listDatasetRuns — List eval execution runs (traces using this dataset)
• getDatasetRun — Get details of a dataset run (eval execution)

Metrics (2 tools)

• getDailyMetrics — Get daily metrics for traces and observations (latency, tokens, scores, cost)
• getUsageSummary — Get usage summary for a time period (total traces, observations, tokens, cost)

Score Configurations (4 tools)

• listScoreConfigs — List all score configurations (custom score names and data types)
• getScoreConfig — Get details of a score configuration including name, data type, and categories
• createScoreConfig — Create a new score configuration for custom scoring metrics
• updateScoreConfig — Update an existing score configuration

Project (1 tool)

• getProject — Get current project details (name, API keys, limits, usage)

Instance Management (8 tools)

• listInstances — List all Langfuse instances with pagination
• getInstance — Get details of a specific instance
• createInstance — Create a new instance
• updateInstance — Update an existing instance
• deleteInstance — Delete an instance
• listInstanceEvents — List events for an instance
• getInstanceStatus — Get health and status information
• configureInstanceSettings — Configure advanced settings

Organization Projects (5 tools)

• listOrganizationProjects — List all projects within an organization
• getOrganizationProject — Get details of a specific project
• createOrganizationProject — Create a new project in an organization
• updateOrganizationProject — Update an existing project
• deleteOrganizationProject — Delete a project from an organization

Organization API Keys (3 tools)

• listOrganizationApiKeys — List all API keys for an organization
• createOrganizationApiKey — Create a new API key
• deleteOrganizationApiKey — Delete an API key

Organization Memberships (6 tools)

• listOrganizationMembers — List all members of an organization
• addOrganizationMember — Add a new member to an organization
• updateOrganizationMember — Update a member's role or permissions
• removeOrganizationMember — Remove a member from an organization
• listProjectMembers — List members with access to a specific project
• updateProjectMember — Update a member's access level for a project

Setup

Environment Variables

# Public API (Required for Prompt Management, Traces, Observations, Scores, Sessions, etc.)
LANGFUSE_PUBLIC_KEY=pk-lf-...
LANGFUSE_SECRET_KEY=sk-lf-...

# Instance/Organization Management APIs (Required for Instance and Organization tools)
LANGFUSE_ADMIN_API_KEY=adm-lf-...

# Optional: Organization-specific API key for additional org-level tools
LANGFUSE_ORG_API_KEY=org-lf-...

# Optional: API Server (defaults to https://cloud.langfuse.com)
LANGFUSE_HOST=https://cloud.langfuse.com
# US region: https://us.cloud.langfuse.com

Usage with Claude Code (via npx)

Add to your MCP configuration (.mcp.json or ~/.claude/settings.json):

{
  "mcpServers": {
    "langfuse": {
      "command": "npx",
      "args": ["-y", "@jurislm/langfuse-mcp@latest"],
      "env": {
        "LANGFUSE_PUBLIC_KEY": "pk-lf-...",
        "LANGFUSE_SECRET_KEY": "sk-lf-...",
        "LANGFUSE_HOST": "https://cloud.langfuse.com"
      }
    }
  }
}

Note: Runtime requires only Node.js >= 18 — the published binary runs under plain node, so npx works without any extra tooling. Bun is needed only for local development (see Development), never to run this server.

Usage with Claude Code Plugin (jurislm-tools)

If you use the jurislm-tools Claude Code plugin, jt:langfuse is included:

/plugin marketplace update jurislm-tools

Then set environment variables in ~/.zshenv:

export LANGFUSE_PUBLIC_KEY=pk-lf-...
export LANGFUSE_SECRET_KEY=sk-lf-...
export LANGFUSE_HOST=https://us.cloud.langfuse.com

Development

bun install
bun run dev        # Run locally (stdio transport)
bun run build      # Build to dist/
bun run typecheck  # TypeScript check
bun run lint       # ESLint (max-warnings=0)

License

MIT