Skip to main content
Guided onboarding is the fastest way to go from zero to a scored run when you are working through an AI coding agent. The agent introspects your organization, picks a published benchmark, configures a run, drives a rollout, and reads back results.

​
The flow

  1. Connect. The agent authenticates with a Verial API key or Solver key. See Authentication.
  2. Discover. The agent lists published benchmarks (GET /benchmarks?visibility=Public), inspects environments, and summarizes what is available.
  3. Pick a benchmark. You confirm a target (for example fax-referral@1). The agent reads the tasks and criteria.
  4. Start a run. The agent calls POST /v1/benchmark-runs, stores the returned bearer token, and walks through each task run.
  5. Drive the rollout. The agent calls your agent-under-test through the sandbox endpoints (FHIR, HL7, files, portal).
  6. Read results. The agent fetches task run verdicts, per-criterion scores, and evidence, then explains what happened.

​
Using MCP

The Verial MCP server exposes every resource as a structured tool. MCP clients can discover the tool surface and execute the onboarding flow without any custom code. See MCP Setup to connect Claude Code, Cursor, or ChatGPT Desktop.

​
Using the CLI

If your coding agent has a shell tool, the Verial CLI is enough: 
verial benchmarks list --json
verial benchmark-runs create --benchmark fax-referral@1 --json
# ... drive rollout endpoints ...
verial benchmark-runs get $RUN_ID --json
A dedicated /onboard slash command that orchestrates the full flow in one step is on the roadmap. In the meantime, point your agent at this page and at Running a Benchmark.

​
Next Steps

Agent Skills

Install Verial expertise into your AI coding agent.

MCP Server

The richest integration for agentic workflows.