Skip to main content
Colter Test

What happens when
agents try to buy?

Colter sends AI shopping personas to your store — they browse, add to cart, and attempt checkout. See exactly where they succeed and where they get stuck.

Start a testView documentation
colter test
$colter test https://my-store.com --browser
Pre-check: WebMCP tools detected (4 tools)
Launching Chrome DevTools MCP...
DISCOVER ✓ Found 4 WebMCP tools
BROWSE ✓ search-products returned 12 results
EVALUATE ✓ Product page loaded, price visible
CART ✓ add-to-cart tool succeeded
CHECKOUT ✗ begin-checkout redirected to login
Score: 78/100 — 4/5 stages passed
Screenshots saved to ./colter-test-screenshots/
How it works

LLM personas with browser verification

Persona Testing

10 AI personas

LLM personas evaluate your store via HTTP — discovery, product info, checkout, edge cases, and security across GPT, Claude, and Gemini.

  • Personas8 shopping archetypes + 2 specialist (security, edge case)
  • ModelsGPT-5, Claude Sonnet, Gemini — cross-model coverage
  • OutputPer-persona scores, journey reports, remediation recommendations

Browser Verification

Included

When your store exposes WebMCP tools, Colter drives a real Chrome browser through the full journey — with screenshots at every step to prove it works.

  • EngineChrome DevTools MCP — real browser, real sessions
  • RequiresWebMCP tools on your store (run Fix to generate them)
  • OutputJourney screenshots, tool invocation log, quality score
Journey stages

Five stages of an agent shopping flow

Each stage maps to a real step in an AI agent's purchase flow.

JOURNEY FUNNEL

18% conversion
Discover100%
22% left· Intent mismatch
Browse78%
33% left· Missing product schema
Evaluate45%
12% left· Variant selector broken
Cart33%
15% left· Checkout requires login
Checkout18%

18% of agents complete a purchase. The biggest drop happens at Browse → Evaluate — most agents can't parse product schema.

For agents and developers

Works from CLI, API, and MCP

AI agents discover and invoke tools programmatically. Developers get the same access from the command line.

CLI
colter test
$colter test https://my-store.com
$colter test https://my-store.com --browser
$colter test https://my-store.com --persona claude_shopper
$colter test https://my-store.com --json | jq .summary.overall_score
REST API
REST APIbash
# Standard test (HTTP + LLM)
curl -X POST https://agenticcom.ai/api/v1/test \
  -H "Authorization: Bearer col_..." \
  -d '{"url": "https://my-store.com"}'
 
# Results include: personas[], summary,
# journey_reports, remediations[],
# per-model scores, total_cost_usd

Install: npx -y @getcolter/cli · MCP: npx -y @getcolter/cli mcp --admin-tools

FAQ

Common questions

What happens during a test?

10 AI personas attempt real shopping journeys on your store across 3 LLM providers (Claude, GPT, Gemini). Each persona tests a different failure mode.

Will the test affect my store?

No purchases are completed. Personas browse, add to cart, and attempt checkout but stop before payment. Read-only except for cart actions.

How is this different from Check?

Check scans your infrastructure. Test sends actual AI agents to shop. Check tells you what’s missing; Test shows you what breaks.

What do I get in the report?

Per-persona scores, journey narratives, remediation recommendations, and a PDF export. You see exactly where each agent succeeded or got stuck.

Does Test include browser testing?

Yes. When your store supports WebMCP tools, Colter also drives a real Chrome browser through the full purchase journey for verification.

Next step

Monitor what happens in production.

Test validates your store works. Lens watches real agent traffic — who visits, what they try, where they drop off.

Learn about Lens