Langfuse MCP server

What the model can do

Langfuse LLM observability, on Langfuse Cloud or self-hosted: recent observations (generations, spans) with model, cost, latency and errors, cost and token usage summaries, evaluation scores, prompts and their versions, datasets, score configs, annotation queues and comments; can add comments and scores, create prompt versions and move their labels, create datasets and dataset items, create score configs and annotation queues and queue items, and delete scores, traces and dataset items.

Reads

• langfuse_list_observationsRecent Langfuse observations (LLM generations, spans, events, tool calls): trace, name, type, level, status message, start and end, latency, model…
• langfuse_usageAggregated Langfuse metrics over observations: cost, tokens, latency or counts, optionally grouped by one dimension (model, observation name…
• langfuse_list_scoresLangfuse evaluation scores (from annotators, LLM judges or the API): name, value, data type, source, when, and what it scores (trace, observation…
• langfuse_list_promptsPrompts managed in Langfuse: name, text or chat, versions, labels (production, staging…), tags and last update.
• langfuse_get_promptOne Langfuse prompt's text (a chat prompt as role: content lines), version, labels, tags, commit message and config.
• langfuse_list_datasetsLangfuse datasets (collections of test cases for experiments): name, description, created and updated.
• langfuse_list_commentsComments left in Langfuse, optionally on one trace, observation, session or prompt (give both object_type and object_id).
• langfuse_list_score_configsLangfuse score configs (the defined scores annotators and queues use): id, name, data type, range or categories, whether archived.
• langfuse_list_annotation_queuesLangfuse annotation queues (human review lists): id, name, description, score config ids and dates.

Acts, when you ask

These change something in Langfuse. The model is told to use them only when you ask for exactly that, and each one is granted by name: per team on a shared connection, by the owner on a personal one.

• langfuse_create_commentAdds a comment (markdown, up to 5000 characters) to a Langfuse trace, observation, session or prompt in this connection's project.
• langfuse_create_scoreAdds an evaluation score (source API) to a Langfuse trace, observation or session: a name and a value, numeric, boolean (1 or 0)…
• langfuse_create_promptCreates a new version of a Langfuse prompt (a new prompt when the name is new): text, or chat messages with optional placeholders.
• langfuse_set_prompt_labelsSets the labels on one version of a Langfuse prompt.
• langfuse_create_datasetCreates a Langfuse dataset (a collection of test cases for experiments) with a name and optional description.
• langfuse_create_dataset_itemAdds a new item (a test case: input and optional expected output, as text or a JSON object) to a Langfuse dataset…
• langfuse_delete_dataset_itemDeletes a Langfuse dataset item and every experiment run result recorded against it.
• langfuse_create_score_configCreates a Langfuse score config, a defined score (numeric with an optional range, boolean, categorical with labelled values…
• langfuse_create_annotation_queueCreates a Langfuse annotation queue (a human review list) scored with the given score configs.
• langfuse_add_queue_itemAdds a trace, observation or session to a Langfuse annotation queue for human review, as pending.
• langfuse_update_queue_itemMarks an item in a Langfuse annotation queue pending or completed.
• langfuse_delete_scoreDeletes one Langfuse score (on a trace, observation or session).
• langfuse_delete_traceDeletes one Langfuse trace with its observations and scores.
• langfuse_delete_tracesDeletes up to 50 Langfuse traces named by id, with their observations and scores.

Connect it

First connect Langfuse on Dashboard → Connections, then add Memtro to your client once. The Langfuse tools appear alongside the tools of everything else you have connected; nothing is configured per connector.

claude plugin marketplace add memtro-org/claude-memtro
claude plugin install memtro@memtro --config api_key=sk-memtro-...

{
  "mcpServers": {
    "memtro": {
      "url": "https://platform.memtro.com/api/mcp",
      "headers": {
        "Authorization": "Bearer sk-memtro-..."
      }
    }
  }
}

{
  "mcp": {
    "memtro": {
      "type": "remote",
      "url": "https://platform.memtro.com/api/mcp",
      "headers": {
        "Authorization": "Bearer sk-memtro-..."
      }
    }
  }
}

Who may use it

The key identifies the person, and the server offers that person the tools they may use everywhere else in Memtro. A shared Langfuse connection is used by the teams an administrator granted it to, tool by tool, on the Permissions page; a personal connection is used by its owner alone, and its tools that act stay off until the owner allows them. See Teams and permissions and Connectors.

Ask it like this

• › What did we spend on LLM calls yesterday, by model?
• › Which generations errored in the last 24 hours?
• › What changed in the production version of the support prompt?
• › Add a score of 0 called correctness to that trace
• › Move the production label to version 7 of the support prompt

Questions

Is this an official Langfuse MCP server?

No. It is Memtro's MCP server with Memtro's Langfuse connector behind it: the connector reads Langfuse through its API with the credentials you connect on Dashboard → Connections, and Langfuse publishes nothing here. The same endpoint carries every other system you connect and Memtro's memory tools.

What does it cost?

Nothing on top of Memtro. One price per organisation by its number of users, from $50 a month for up to 10 users after a 7-day free trial, with every connector, the MCP server, jobs and memory included. You bring your own model keys.

Can it change anything in Langfuse?

Only when you ask, and only what it has been granted: add a comment to a trace, observation, session or prompt; add a score to a trace, observation or session; create a new prompt version (labels such as production change what apps serve); move labels such as production onto a prompt version; create a dataset; add an item to a dataset; delete a dataset item and its run results (permanent); create a score config; create an annotation queue; add a trace, observation or session to an annotation queue; mark an annotation queue item pending or completed; delete a score (permanent); delete a trace (permanent); delete up to 50 traces by id (permanent). Every other tool reads, searches or lists. An administrator grants each acting tool by name per team, and a personal connection acts only after its owner allows it.

Which clients work?

Any MCP client that speaks Streamable HTTP: Claude Code, Cursor, Claude Desktop, OpenCode, Codex, Zed, Windsurf, Cline and the Anthropic and OpenAI SDKs. There is no local process to run; the server is remote and authenticated with your key.

Other MCP servers · Engineering and operations

• GitHub MCP server
• AWS CloudWatch MCP server
• Sentry MCP server
• ServiceNow MCP server
• Okta MCP server
• Twilio MCP server