- ANALYSIS
- TRENDS
- DETECTION
- DECISION
Short answer
How do I check if my model changes improved accuracy?
Your AI queries Braintrust to pull experiment scores and project analytics. It compares these metrics to show you exactly how recent changes impacted behavior. Here is how it works.
The outcomes
What you get from the data
Your AI turns Braintrust metrics into clear answers.
ANALYSIS
Compare experiment scores
The AI pulls data from your latest runs to show how different versions perform against each other.
TRENDS
Track project history
Your agent looks at project analytics to see if performance is trending up or down over time.
DETECTION
Spot behavior shifts
The AI identifies when a specific model change causes a drop in accuracy or a change in output style.
DECISION
Human-led deployment
The AI presents the findings, but you make the final call on which model to push to production.
The workflow
What your AI does when the data arrives.
The AI acts as a bridge between your Braintrust dashboard and your chat interface.
Locate projects
The AI finds the specific Braintrust projects you want to investigate.
list_projectsFetch experiment scores
It pulls the specific metrics and scores from your recent test runs.
list_experimentsCompare versions
The agent looks at different experiment results to find performance deltas.
list_experimentsReport findings
Your AI summarizes the project analytics to explain how behavior has changed.
list_projects
Try it
Copy these to start.
Use these prompts to begin your analysis.
Starting points
Replace the project names and experiment IDs with your actual Braintrust names.
Braintrust Connector
You're all set. Choose your MCP client and follow the setup instructions.
https://edge.vinkius.com/vk_preview_cr8T7CpPGy155EtBgklGtlDgFsTWLRLYfzKoHIOF/mcpClaude Desktop
Follow the steps below to connect in seconds.
- 1In Claude Desktop, open Settings → Connectors.
- 2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
- 3Click Add and start a new chat — Braintrust capabilities are ready to use.
{
"mcpServers": {
"braintrust-mcp": {
"url": "https://edge.vinkius.com/vk_preview_cr8T7CpPGy155EtBgklGtlDgFsTWLRLYfzKoHIOF/mcp"
}
}
}List all my Braintrust projects so I can see what's available.
Show me the scores for the 'v2-test-run' experiment in my 'Customer-Bot' project.
Compare the accuracy scores between my last three experiments.
Did the recent changes in the 'Search-Optimization' project improve my model's performance?
Claude
ChatGPT
Cursor
VS Code
Windsurf
Claude Code
JetBrains
Cline
Start here
Connect Braintrust once, then ask.
Connect your account in one step. Your credentials stay encrypted on our side, and you can immediately start asking your AI about your model's performance.
Connect Braintrust to your AIFAQ
How this task behaves
- 01
Can the AI change my experiment settings?
No. The AI can only read project and experiment data; it cannot modify your Braintrust configurations.
- 02
Can the AI delete my test runs?
No. This MCP is read-only for your experiments and projects.
- 03
How does the AI know which project to look at?
It uses the list_projects tool to see your available projects, then you can tell it which one to focus on.
- 04
Does this work with any AI client?
Yes, it works with any MCP-compatible client like Claude, Cursor, or Windsurf.
- 05
Can the AI create new experiments for me?
No. The AI can only list and query existing experiments.
More questions about Braintrust? See everything the Braintrust Connector can do
Connect Braintrust to Claude, Cursor, ChatGPT & more
