What is agent readiness?

How much of a shopping journey an AI agent can complete on your store: finding products, reading prices and stock, adding to a basket, reaching checkout, paying, and starting a return. tragent scores it out of 100 by driving a real agent through each stage.

Why can't ChatGPT buy from my store?

Usually one of three things: bot management refuses the agent, your product data is not machine-readable, or checkout needs a gesture only a person can make. A scan names which one it is and where it happened.

Does tragent place real orders?

No. By default a journey stops before payment. On a staging store you can let it complete checkout against a payment processor's test endpoints, where no money moves and the processors' live hosts stay blocked.

Which platforms does it work with?

Any store with a public URL. Shopify connects with an API token and configures itself; WooCommerce, Magento, BigCommerce, commercetools and custom stacks are tested the same way once you say where your feed and checkout are.

Sample AI shopping readiness82 / 100

Agentic commerce testing and AI shopping agent readinessAI agents are trying to buy from your store. Most can’t finish.

tragent tests whether ChatGPT, Gemini, Perplexity and browser agents can find, price and buy from your store, shows the exact step where they fail and what it costs, and blocks the regression in CI.

See a real example: the sample store

tragent/Northwind SupplyAll storesLast 7 days

Home

Northwind Storefront

https://staging.northwind.example · Shopify · staging

Start a test

AI shopping readiness

0/ 100

Not measured

0 of 9 stages measured

Revenue at riskEstimate

—/mo

From the stages agents cannot get past

Latest test

In progressAI shopping test2 min ago

staging · 9be77d1

Open the test

Find productsNot measured
Read the detailsNot measured
See the priceNot measured
Check stockNot measured
Add to basketNot measured
Reach checkoutNot measured
PayNot measured
Return an orderNot measured
Sign inNot measured

Fix this first · 3 issues, by revenue at risk

1
HighPay Where agents first get stuck
Delegated payment is not accepted
Affects the assistant-style browser agent · typical effort: engineering work
2
MediumReturn an order
Returns need a CAPTCHA only a person can pass
3
LowReach checkout
Agent authorisation header rejected on 2 of 5 tries

tragent / AI shopping readiness

Failing after 42s

Required

Pay stopped passing on this pull request

Checkout success rate · assistant-style agents7d

94%Mon

91%Tue

67%Wed

Fell from 94% to 67% after release 9be77d1

Change since the reference result1 stage regressed

PayPassed→Failed

Reason Delegated payment not accepted.

prev a41f2c0·now 9be77d1

What every run coversStagesdiscovery → refundAgent stylesbrowser · API-firstDiagnosiscause code + fixProtocol checksUCP · ACP · AP2 · MCP · WebMCP · A2ACIGitHub Actions · GitLab CIEvery step tracedscreenshot + DOM

Product

One platform for AI shopping readiness. Tests, release checks and monitoring, around the clock.

AI shopping tests
Simulated agents with different styles run goal-driven journeys against your stack.
Agent styles2 enabled
- passAssistant-style browser agentdrives the UI like ChatGPT, Gemini or Claude5 runs
- passAPI-first agentprefers structured endpoints, falls back to UI5 runs
Readiness scoring
Every journey is scored against nine stages. The diagnosis — where and why — is the product.
StagesTest plan retail-core · 2 agent styles
- passFind productsfound 3 matching products via search
- passRead the detailsschema.org Product + Offer present
- passSee the price$89.00 USD matches visible price
- passCheck stockInStock · size 10 available
- passAdd to basket1 item · subtotal $89.00
- warningReach checkoutauthorisation header rejected on 2 of 5 tries
- failPaydelegated payment not accepted
- failReturn an orderreturns need a visual CAPTCHA
- warningSign inno delegated sign-in available
Release checks in CI
Run AI shopping tests on every pull request and block a release that makes agents worse off.
tragent-botcommented 2 minutes agoblocked
AI shopping readiness: 1 stage regressed
readiness 88 → 82
retail-core · 9 stages · 2 agent styles- Pay Passed+ Pay Failed delegated payment not accepted Return an order Failed (unchanged)
Release check: min_score 75 · fail_on regression → blocked
Monitoring
Scheduled checks and real agent traffic, per agent type, with alerts the moment a journey breaks.
Checkout success rate · assistant-style agents7d
94%Mon
91%Tue
67%Wed
Fell from 94% to 67% after release 9be77d1

One rule, every market

One rule blocks the deploy. Agents shop from everywhere you ship to.

whenA run finishes onmain
ifReadiness is under75or any stageregresses
thenBlock the deploy
andComment on the pull request

The gate as it reads in a suite. The thresholds are yours to set.

Illustrative. Agentic checkout is live in North America and Europe first — these are markets, not measurements.

How it works

Point, test, protect. From a URL to a release check in one afternoon.

1
Point tragent at a URL
Give us a staging or production store and, optionally, a test account. No SDK required to start.
```
$ tragent test \
    --target https://staging.shop.example \
    --suite retail-core.yaml
```
2
Agents run real journeys
Assistant-style browser agents and API-first agents try to discover, compare, add to cart and buy.
```
persona  assistant-browser
goal     "Buy a black Nike shoe, size 10"
steps    14 · tokens 21,480 · 38s
```
3
Block the release
Every test becomes a readiness report with the exact step and reason each agent failed. In CI, a stage that gets worse fails the build.
```
readiness  82 / 100
gate       min_score 75 · fail_on regression
result     FAIL — Pay regressed
```

CLI and npm package

One command in your terminal. The same check on every pull request.

The free scan runs on your machine with no account and no key, so it works against a dev server before anything ships. With an API key, the same CLI runs AI shopping tests in CI and fails the build when a stage gets worse.

Terminal

$ npx @tragent/cli scan https://your-store.com

Prebuilt for macOS, Linux and Windows on x64 and arm64, checked against SHA-256 checksums.
Exit code 0 passed, 1 the release check failed, 2 nothing could be measured.

CLI reference Set it up in CI

Terminal

$ npx @tragent/cli scan https://staging.northwind.example

AI shopping readiness — https://staging.northwind.example
90 / 100 · warned · 3 findings
scored 4 of 9 stages · assistant-browser · 14s
warned: Your product data does not say whether it is in stock (medium, at Check stock)

  ✓ Find products    100 
  ✓ Read the details 100 
  ✓ See the price    100 
  ⚠ Check stock       50 2 of 24 offers have no availability.
  – Add to basket      – not attempted by this run
  – Reach checkout     – not attempted by this run
  – Pay                – not scored; no payment method agents can use
  – Return an order    – not attempted by this run
  – Sign in            – not attempted by this run

A sample run against the sample store. The scan scores the four stages it can read; basket, checkout, payment, returns and sign-in need an AI shopping test, which is where the dashboard's 82 comes from.

npm
npm i -g @tragent/cli
View on npm
GitHub Actions
uses: tragent-io/cli@v1
Action reference
GitLab CI
remote: …/tragent-io/cli/v1/gitlab/.tragent.yml
Template reference

Where agents fail

The failures nobody files a ticket for. Agents do not complain. They leave.

Find products: failed
Read the details: failed
See the price: passed
Check stock: passed
Add to basket: passed
Reach checkout: failed
Pay: passed
Return an order: failed
Sign in: failed

BOT_RULE_BLOCKS_AGENTFind products
A broad bot-management rule returns 403 to declared commerce agents alongside the scrapers it meant to stop.
UCP_PROFILE_INVALIDFind products
The /.well-known/ucp profile lists endpoints that 404 or capabilities the API does not implement.
SCHEMA_OFFER_MISSINGRead the details
Name and image are marked up; price, currency and availability are not. Agents guess or skip the product.
AUTH_HEADER_REJECTEDReach checkout
A WAF or middleware strips the header a verified agent sends, so checkout 401s while the browser path passes.
CAPTCHA_ON_FLOWReturn an order
Returns and refunds sit behind a challenge no agent can pass, so post-purchase journeys dead-end.
NO_DELEGATED_LOGINSign in
Reorders and account actions need a full interactive login. An agent acting for a customer has no sanctioned way in.

Integrations

Works with what you already run. Keep your platform, CDN and payment stack.

CI: GitHub ActionsGitLab CIAny CI via CLI
Commerce: Shopify — one-click WooCommerce Magento BigCommerce commercetools Custom
Agent standards: ChatGPT checkout (ACP)Shared catalogue (UCP)Payment authorisation (AP2)Direct tool access (MCP)Browser agent checkout (WebMCP)Agent discovery (A2A)
Alerts: Slack, per projectEmail, per projectPagerDuty, per ruleWebhooks, per rule

Why it matters

Built for teams where agent failures cost money. A silent failure has no ticket, no error and a real invoice.

IllustrationWorked example — your own figures go in the revenue model

10,000,000
monthly visitors: × 5%
agent-driven sessions: × 2%
conversion: =10,000
transactions a month at risk

Built forE-commerce·Travel·Hotels·Airlines·Food delivery·Ticketing·Marketplaces·B2B procurement·Insurance·Banking·SaaS

Find out before your customers' agents do. Start with a free scan, look around the sample store, or create an account and test your own staging site.

View the demo

Agentic commerce testing and AI shopping agent readinessAI agents are trying to buy from your store. Most can’t finish.

One platform for AI shopping readiness. Tests, release checks and monitoring, around the clock.

AI shopping tests

Readiness scoring

Release checks in CI

Monitoring