Braintrust MCP server

What the model can do

Braintrust LLM evaluation: projects, experiments with average scores and how they moved against a baseline, dataset names, functions, and a project's recent logs with inputs and outputs cut to 300 characters. When asked, comments on an event, creates, renames or deletes projects, experiments and datasets, adds dataset examples and log events, saves a chat prompt, runs a function, and adds review scores and tags.

Reads

• braintrust_list_projectsBraintrust projects, newest first: id, name, description and when created.
• braintrust_list_experimentsExperiments in a Braintrust project, newest first: id, name, description, created, commit, dataset and tags.
• braintrust_experiment_summaryOne Braintrust experiment's average scores (0 to 1) and metrics (duration, tokens, cost) with how each moved against a comparison experiment…
• braintrust_list_datasetsDatasets in a Braintrust project by name: id, name, description, created and tags.
• braintrust_project_logsA Braintrust project's most recent logged traces, newest first: span name, model, input, output, expected and error each cut to 300 characters…
• braintrust_list_functionsFunctions (prompts, scorers, tools, tasks) in a Braintrust project: id, name, slug, type, description and when created.

Acts, when you ask

These change something in Braintrust. 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.

• braintrust_add_commentAdd a comment to one event in Braintrust, either a project log event (project_id) or an experiment event (experiment_id)…
• braintrust_create_projectCreates a Braintrust project, visible to everyone in the organisation with project access.
• braintrust_update_projectRenames a Braintrust project or changes its description.
• braintrust_delete_projectDeletes a Braintrust project with all its experiments, datasets, logs, prompts and scores, for everyone in the organisation.
• braintrust_create_experimentCreates an empty experiment in a Braintrust project, optionally linked to a dataset and a baseline experiment, for an eval run to log into.
• braintrust_update_experimentRenames a Braintrust experiment or changes its description, tags, baseline or linked dataset.
• braintrust_delete_experimentDeletes a Braintrust experiment and its results for everyone in the organisation.
• braintrust_create_datasetCreates an empty dataset in a Braintrust project, to hold eval examples.
• braintrust_add_dataset_rowsAdds up to 50 examples to a Braintrust dataset, each an input with optionally its expected output, metadata and tags.
• braintrust_log_eventsWrites up to 50 events into a Braintrust project's logs, each with an input, output, optionally expected value, error, scores (0 to 1)…
• braintrust_create_promptSaves a chat prompt in a Braintrust project under a slug, with its messages and model.
• braintrust_run_functionRuns a Braintrust function (a saved prompt, scorer or tool) on one input and returns its result, cut to 4,000 characters.
• braintrust_create_scoreAdds a human-review score to a Braintrust project: a slider (0 to 1), a categorical score with named values, or free-form text…
• braintrust_create_tagAdds a tag to a Braintrust project's tag list, for marking logs, experiments and dataset rows.
• braintrust_delete_datasetDeletes a Braintrust dataset and its rows for everyone in the organisation; experiments that used it lose the link.

Connect it

First connect Braintrust on Dashboard → Connections, then add Memtro to your client once. The Braintrust 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 Braintrust 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

• › Did last night's eval regress?
• › Show recent low-scoring answers in the support bot project.
• › How did cost and tokens move in the latest experiment against its baseline?
• › Add these three failed support answers to the refunds dataset with the expected replies.
• › Add a comment to that log event saying retrieval missed the refunds policy.

Questions

Is this an official Braintrust MCP server?

No. It is Memtro's MCP server with Memtro's Braintrust connector behind it: the connector reads Braintrust through its API with the credentials you connect on Dashboard → Connections, and Braintrust 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 Braintrust?

Only when you ask, and only what it has been granted: add a comment to a logged or experiment event; create a project; rename a project or change its description; delete a project with everything in it; create an experiment, linked to a dataset or baseline; rename an experiment or change its description, tags, baseline or dataset; delete an experiment and its results; create a dataset; add examples to a dataset; write events into a project's logs; save a chat prompt, or a new version of one, under a slug; run a saved prompt, scorer or tool on an input, spending model tokens; add a slider, categorical or free-form review score to a project; add a tag to a project; delete a dataset. 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