Skip to main content

Quickstart

Get your first AI agent experience report in under 5 minutes.

1. Sign up

Go to 2027.dev/evals and sign in with your work email. Your organization is automatically detected from your email domain.

2. Scan your domain

Click Scan and enter your product’s URL. Our agent will browse your docs site and generate:
  • A provider profile (name, category, tags)
  • An eval prompt that tests your core workflow
You can edit both before saving.

3. Run your first eval

Once the prompt is saved, click Request Eval. The eval takes 5-15 minutes to complete. You’ll get a Slack notification when it’s done.

4. Review your report

Your report includes:
  • Overall grade (A+ to F) and score (0-100)
  • Dimension breakdown — documentation, API design, error handling, etc.
  • Friction points — specific moments where the agent struggled
  • Session trace — full replay of the agent’s conversation
  • Screenshots — what the agent saw in the browser
  • Recommendations — prioritized improvements ranked by effort

5. Get your API key

To use the MCP server or API programmatically:
  1. Go to SettingsAPI Keys in the dashboard
  2. Click Create Key and copy it (starts with evals_)
  3. Use it in the Authorization: Bearer evals_YOUR_API_KEY header
No npm package install is needed — the MCP server is HTTP-based. See MCP Server setup for config examples.

What’s next?

Connect MCP

Query reports from Claude Code or Cursor

GitHub Integration

Auto-eval on every PR

Understanding Scores

How scoring works