Build custom agent evals via our MCP.We handle the infrastructure.
Create and maintain evals through your own AI agent, using your team’s domain expertise. Bridge provides the tooling, dataset storage, versioning and runtime to execute them.
Create and maintain evals with
Cursor
Codexor any MCP-compatible agentYour AI assistant
Bridge MCP“Test that our payments agent verifies new bank details before paying.”
Using your policy + agent context
Saved & run on Bridge
Failed: agent paid before verification.
Keep the observability tools you already use.
Bring real cases from your traces into evals. Keep your monitoring, your harness and your domain expertise.
Braintrust
and other observability toolsFrom a failed run
to a repeatable test.
Start with a real conversation, add the judgment only your team has, then test a change. This example follows a payments agent that skipped verification.
Define the test: Eval saved · v3
Your AI assistant Bridge MCP connected
YouOur agent paid before verifying new bank details. Create 24 tests using our policy and reviewer exception.
Your assistantI’ve used your policy and payment tools to define the scenarios and pass criteria.
Context: Braintrust trace · your payment policy · agent tools
bridge.save_evalMCP details
{
"name": "supplier_bank_change",
"source_trace": "braintrust/payment_017",
"type": "multi_turn",
"cases": 24,
"rubric": [
"hold_payment",
"verify_details",
"record_evidence"
],
"exception": "authorized_reviewer_approval"
}On Bridge
Eval saved · v3Supplier bank-detail changes
Multi-turn simulation · 24 cases · eval v3
Your team’s pass criteria
- Hold the payment
- Verify the new bank details
- Record the evidence
Exception: an authorized reviewer can approve the change.
Your expertise. Your test structure.
A simple response check or a conversation with branching rules and multiple judges. Your team defines the scenario and pass criteria through its own AI assistant. Bridge stores and versions the complete eval definition.
Your stack stays yours.
Keep your agent harness and observability tools. Bridge handles eval execution and performance history, alongside your existing stack. Engineers connect the runtime; domain experts help build and maintain the tests.
Your team defines good.
We keep the evals running.
Behind the conversation is shared eval infrastructure. Store your team’s definitions and datasets, execute them through your configured runtime, and keep a history of behavior as the product changes.
Keep your agent, your context and your custom grading logic.
Define your own evals
Single-turn checks, multi-turn scenarios, custom rubrics or branching LLM judges. Store the definition your task needs.
Store and version everything
Keep eval definitions, datasets, rubrics and judge configurations together, with the history behind every change.
Run and track performance
Execute evals through your configured runtime. Compare runs and track performance over time against versioned definitions.