Work with Braintrust traces and evals inside Superconductor coding agents
Connect Braintrust to Superconductor and your coding agents can now look into production traces, write scorers, and run evals before and after they change a prompt or model. Superconductor runs that work in a shared session, so a teammate can review the traces and eval results behind a change alongside the code.
Superconductor lets you and your team chat with coding agents such as Claude Code, Codex, Pi, and more in shared, secure cloud sessions.
With our Braintrust integration, you can
- Find the traces behind a bad answer and spot what they share
- Write and test a scorer for a new failure mode
- Run an eval to check a prompt change before it ships
Try asking
- Add this week's ten worst support bot answers to our "regressions" dataset in Braintrust, so future evals catch them.
- Set up a Braintrust alert that tells us when the online score for our support bot drops below 0.8.
- Look for recurring patterns in this week's Braintrust traces and report back. Don't create or change anything.
Technical details
- Agents can
- Query logs, experiments, and datasets; manage prompts, scorers, patterns, alerts, and automations; run evals
- Access
- The projects the connected Braintrust account, API key, or service token can reach
- Needs
- Sign in with Braintrust, or a Braintrust API key or service token
- Server
- Braintrust's official hosted MCP server
Questions
- Does it use Braintrust's official MCP server?
- Yes. Agents connect to Braintrust's hosted MCP server at
api.braintrust.dev. Sign in with Braintrust, or paste a Braintrust API key or service token. - Does it work with the EU data plane or self-hosted Braintrust?
- The built-in connector uses Braintrust's US endpoint. For the EU data plane or a self-hosted deployment, add your organization's MCP URL as a custom MCP server.
- Can agents change my Braintrust data?
- Yes. Agents can create prompts, scorers, alerts, and dataset rows, and run evals, so tell them when they should only look.