Skip to main content
The Braintrust MCP Server gives AI agents within Caylex direct, programmatic access to both Braintrust and OpenAI documentation alongside live Braintrust project data, enabling end-to-end workflows for experimentation, observability, and debugging. Agents can search documentation, resolve and list Braintrust objects (such as experiments, datasets, and logs), infer data schemas, run SQL queries against Braintrust data, summarize experiment results, and generate shareable permalinks — all without leaving the Caylex environment. This makes it well-suited for teams that need their AI agents to autonomously analyze evaluation runs, triage production issues, or assist developers in integrating and extending Braintrust-powered pipelines. By bridging documentation lookup with live data access, the server supports both exploratory and operational use cases in a single connection.
Prerequisites:
  • A Braintrust account with at least one project is required to access live data tools such as experiment summaries, SQL queries, and object resolution.
  • A Braintrust API key with read access to the relevant projects must be available to authenticate the server connection in Caylex.
  • Teams on a self-hosted Braintrust deployment must know their custom data plane URL to correctly configure the server.
Important API details:
  • A valid Braintrust API key is required to access live project data such as experiments, datasets, and logs; documentation search does not require authentication.
  • SQL queries are executed against Braintrust’s internal data plane — teams using a self-hosted Braintrust data plane must configure the server to point to their custom data plane endpoint rather than the default cloud endpoint.
  • Object resolution and listing tools operate within the scope of the authenticated user’s Braintrust organization and project permissions.
  • Permalink generation produces URLs tied to the authenticated user’s Braintrust organization, so generated links will only be accessible to users with appropriate access to that organization.

​
Server Details

PropertyValue
TransportStreamable HTTP
Server TypeExternal
Endpointhttps://api.braintrust.dev/mcp
CategoriesAPI Tools & Services, Development & DevOps

​
Authentication

This server supports the following authentication methods:

​
OAuth

Scopes: mcp During the server onboarding flow, you will be prompted to complete the OAuth flow to grant access. See the Braintrust Authentication guide for step-by-step credential configuration.

​
Header Authentication

This server authenticates via HTTP headers. During the server onboarding flow, you will be prompted to confirm the required headers. Note that you do not provide the values for the headers during the server on-boarding. Header values are provided by the users during the Auth Link flow.

​
Server Configuration

The following parameters can be set in the server URL when adding this server to your Caylex project:
ParameterTypeDescriptionExample
api_urlqueryOverride the default Braintrust API/data plane URL for organizations using a self-hosted Braintrust deployment. Without this, the server defaults to the Braintrust cloud data plane.https://your-braintrust-instance.example.com

​
Getting Started

1

Add the server

Navigate to the Server Library and click on the New Server button. Find Braintrust in the Caylex Catalog.
2

Server Onboarding flow

Go through the server onboarding flow.
3

Use in a project

Add the server to a project by configuring project connections. Its tools are now available to any agents connected to that project.
https://d338mlbnszozgc.cloudfront.net/logos/datadog.svg

Datadog

https://d338mlbnszozgc.cloudfront.net/logos/posthog.svg

Posthog

​
References

https://www.braintrust.dev/docs/reference/mcp

https://www.braintrust.dev/docs/kb/configure-mcp-server-with-self-hosted-data-plane