Add to your Claude Desktop configuration file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "langfuse": {
      "command": "npx",
      "args": ["-y", "langfuse-mcp-extended"],
      "env": {
        "LANGFUSE_PUBLIC_KEY": "pk-lf-...",
        "LANGFUSE_SECRET_KEY": "sk-lf-..."
      }
    }
  }
}

For EU Cloud or self-hosted, add LANGFUSE_BASE_URL:

{
  "mcpServers": {
    "langfuse": {
      "command": "npx",
      "args": ["-y", "langfuse-mcp-extended"],
      "env": {
        "LANGFUSE_PUBLIC_KEY": "pk-lf-...",
        "LANGFUSE_SECRET_KEY": "sk-lf-...",
        "LANGFUSE_BASE_URL": "https://eu.cloud.langfuse.com"
      }
    }
  }
}

Option 1: One-liner with environment variables

claude mcp add langfuse -e LANGFUSE_PUBLIC_KEY=pk-lf-xxx -e LANGFUSE_SECRET_KEY=sk-lf-xxx -- npx -y langfuse-mcp-extended

Option 2: Add to project's .mcp.json

{
  "mcpServers": {
    "langfuse": {
      "command": "npx",
      "args": ["-y", "langfuse-mcp-extended"],
      "env": {
        "LANGFUSE_PUBLIC_KEY": "pk-lf-...",
        "LANGFUSE_SECRET_KEY": "sk-lf-..."
      }
    }
  }
}

Option 3: Use shell environment variables

claude mcp add langfuse -- npx -y langfuse-mcp-extended

Then set in your shell profile (.bashrc, .zshrc, etc.):

export LANGFUSE_PUBLIC_KEY="pk-lf-..."
export LANGFUSE_SECRET_KEY="sk-lf-..."

Add to your Cursor MCP configuration:

• Project-specific: .cursor/mcp.json in your project directory
• Global: ~/.cursor/mcp.json in your home directory

{
  "mcpServers": {
    "langfuse": {
      "command": "npx",
      "args": ["-y", "langfuse-mcp-extended"],
      "env": {
        "LANGFUSE_PUBLIC_KEY": "pk-lf-...",
        "LANGFUSE_SECRET_KEY": "sk-lf-..."
      }
    }
  }
}

Or add via UI: File → Preferences → Cursor Settings → MCP

Add to .vscode/mcp.json in your workspace:

{
  "mcpServers": {
    "langfuse": {
      "command": "npx",
      "args": ["-y", "langfuse-mcp-extended"],
      "env": {
        "LANGFUSE_PUBLIC_KEY": "pk-lf-...",
        "LANGFUSE_SECRET_KEY": "sk-lf-..."
      }
    }
  }
}

npm install -g langfuse-mcp-extended

Then use langfuse-mcp-extended as the command instead of npx -y langfuse-mcp-extended.

listTraces

List traces with filtering and pagination.

Inputs:

• page (number, optional): Page number
• limit (number, optional): Items per page
• name (string, optional): Filter by trace name
• userId (string, optional): Filter by user ID
• sessionId (string, optional): Filter by session ID
• tags (string[], optional): Filter by tags (AND logic)
• fromTimestamp (string, optional): Start of time range (ISO 8601)
• toTimestamp (string, optional): End of time range (ISO 8601)
• environment (string, optional): Filter by environment
• orderBy (string, optional): Sort by timestamp, latency, or totalCost

getTrace

Get a specific trace with full details.

Inputs:

• traceId (string, required): The trace ID
• includeIO (boolean, optional): Include input/output fields. Default: false

Returns: Trace with observations and scores.

deleteTrace

Delete a trace.

Inputs:

• traceId (string, required): The trace ID to delete

listObservations

List observations (generations, spans, events) with cursor-based pagination.

Inputs:

• cursor (string, optional): Pagination cursor
• limit (number, optional): Items per page
• traceId (string, optional): Filter by trace ID
• name (string, optional): Filter by observation name
• type (string, optional): Filter by type: GENERATION, SPAN, EVENT
• parentObservationId (string, optional): Filter by parent
• fromStartTime (string, optional): Start of time range (ISO 8601)
• toStartTime (string, optional): End of time range (ISO 8601)
• userId (string, optional): Filter by user ID
• version (string, optional): Filter by version

getObservation

Get a specific observation with all details.

Inputs:

• observationId (string, required): The observation ID
• includeIO (boolean, optional): Include input/output fields. Default: false

Returns: Observation with usage, costs, and timing.

createScore

Create a score for a trace or observation.

Inputs:

• traceId (string, required): ID of the trace to score
• name (string, required): Score name (e.g., accuracy, relevance)
• value (number | string, required): Score value
• observationId (string, optional): ID of specific observation to score
• dataType (string, optional): NUMERIC, CATEGORICAL, or BOOLEAN
• comment (string, optional): Explanation of the score
• configId (string, optional): Score config ID for validation
• id (string, optional): Custom ID for idempotency

listScores

List scores with filtering.

Inputs:

• page, limit (pagination)
• traceId (string, optional): Filter by trace ID
• observationId (string, optional): Filter by observation ID
• name (string, optional): Filter by score name
• source (string, optional): Filter by source: API, ANNOTATION, EVAL
• dataType (string, optional): Filter by data type
• configId (string, optional): Filter by config ID
• fromTimestamp, toTimestamp (string, optional): Time range
• userId (string, optional): Filter by user ID
• operator (string, optional): Comparison operator: <, >, <=, >=, !=, =
• value (number, optional): Value to compare (requires operator)

getScore

Get a specific score.

Inputs:

• scoreId (string, required): The score ID

deleteScore

Delete a score.

Inputs:

• scoreId (string, required): The score ID to delete

createScoreConfig

Create a score configuration template.

Inputs:

• name (string, required): Unique config name
• dataType (string, required): NUMERIC, CATEGORICAL, or BOOLEAN
• minValue (number, optional): Minimum value (NUMERIC only)
• maxValue (number, optional): Maximum value (NUMERIC only)
• categories (array, optional): Category definitions (CATEGORICAL only)
• description (string, optional): Description

listScoreConfigs

List all score configurations.

Inputs:

• page, limit (pagination)

getScoreConfig

Get a specific score configuration.

Inputs:

• configId (string, required): The config ID

createDataset

Create a new dataset.

Inputs:

• name (string, required): Dataset name
• description (string, optional): Description
• metadata (object, optional): Additional metadata

listDatasets

List all datasets.

Inputs:

• page, limit (pagination)

getDataset

Get a dataset by name.

Inputs:

• datasetName (string, required): The dataset name

createDatasetItem

Create or update a dataset item.

Inputs:

• datasetName (string, required): Target dataset name
• input (any, required): Item input data
• expectedOutput (any, optional): Expected output
• metadata (object, optional): Item metadata
• sourceTraceId (string, optional): Source trace ID
• sourceObservationId (string, optional): Source observation ID
• id (string, optional): Custom ID for upsert

listDatasetItems

List items in a dataset.

Inputs:

• datasetName (string, required): Dataset name
• page, limit (pagination)
• sourceTraceId (string, optional): Filter by source trace
• sourceObservationId (string, optional): Filter by source observation

getDatasetItem

Get a specific dataset item.

Inputs:

• datasetItemId (string, required): The item ID

deleteDatasetItem

Delete a dataset item.

Inputs:

• datasetItemId (string, required): The item ID to delete

createDatasetRunItem

Link a trace/observation to a dataset item for evaluation.

Inputs:

• runName (string, required): Name of the dataset run
• datasetItemId (string, required): Dataset item ID
• traceId (string, required): Trace ID
• runDescription (string, optional): Run description
• observationId (string, optional): Observation ID
• metadata (object, optional): Run item metadata

listDatasetRuns

List runs for a dataset.

Inputs:

• datasetName (string, required): Dataset name
• page, limit (pagination)

getDatasetRun

Get a specific dataset run.

Inputs:

• datasetName (string, required): Dataset name
• runName (string, required): Run name

listSessions

List all sessions.

Inputs:

• page, limit (pagination)
• fromTimestamp (string, optional): Start of time range (ISO 8601)
• toTimestamp (string, optional): End of time range (ISO 8601)

getSession

Get a specific session with its traces.

Inputs:

• sessionId (string, required): The session ID
• includeIO (boolean, optional): Include input/output in traces. Default: false

Returns: Session with associated traces.