# On: agent door 43, human door not measured
- Site: on.com
- Scanned: 2026-09-04
- Scan type: Basic scan
- Business type: unknown
- Report: https://www.seconddoor.io/r/on-4592me
- Scanner version: 10
A retail storefront with no machine-readable product data, pricing, or checkout path for agents to complete purchases.
## Scores
| Measure | Score |
| --- | --- |
| Humans | not measured |
| Agents | 43/100 |
| Composite | 43/100 |
Agent door only. The human door could not be measured on this scan. A Door Gap needs both doors, so this scan does not report one.
## What the machine actually did
The agent got through 10 of 17 requests, but could not find a contact route for machines. It came up empty on 6 other checks too. That is enough missing for a machine to give up before it reaches a purchase.
- ok      reach the site at all
- ok      find out whether machines are welcome
- ok      ask whether the door opens for requests carrying the AI crawlers' names
- ok      read the site's own guide for language models
- ok      find out what pages exist
- missing find a contact route for machines
- ok      find a published agent interface
- missing find an agent plugin manifest
- missing find out whether verified agents are recognised
- ok      read the homepage as data rather than as a page
- missing find a machine readable API spec
- missing read the developer documentation
- ok      read the prices without running JavaScript
- ok      read the product catalogue as data
- missing read a product page as data
- missing read the returns and shipping terms
- ok      walk the way a visitor would, to a signup form or a product
## Top issues
1. **The product pages carry no Product structured data, so an assistant must scrape the page to learn a price, and cannot learn availability or an identifier at all.** (commercial_rails)
2. **No returns or shipping policy page reads as text at the usual paths, so an assistant has no terms to quote beside the price.** (machine_fetchable_trust)
3. **No developer documentation is linked from the homepage, in raw HTML or the rendered page, so integration starts with a search engine.** (programmatic_onboarding)
The evidence and the recommended fix for each are on the report page: https://www.seconddoor.io/r/on-4592me
## Files generated from this scan
- **robots.txt rules for AI crawlers** at `/robots.txt`, 10 minutes, no developer needed. Verify: The next scan reads robots.txt and finds every search and agent crawler allowed, none of them kept off product or pricing paths.
- **Product markup skeleton** at `/en-us/shop/womens`, 2 hours, developer. Verify: The next scan reads price, currency and availability in the Product markup of at least one product page.
- **Returns and refunds page** at `/returns`, 30 minutes, no developer needed. Verify: The next scan reads the returns policy page as text without JavaScript and finds it linked from a product page.
- **Shipping and delivery page** at `/shipping`, 30 minutes, no developer needed. Verify: The next scan reads the shipping policy page as text without JavaScript and finds it linked from a product page.
- **Machine interfaces: where you stand** at `notes`, 10 minutes, no developer needed. Verify: The next scan finds the interface you published answering at its well-known path, and that check passes.
- **Agent checkout: where you stand** at `notes`, 10 minutes, no developer needed. Verify: The next scan finds a UCP manifest, a storefront MCP endpoint or another discovery file answering, and the checkout check passes.
- **security.txt** at `/.well-known/security.txt`, 10 minutes, no developer needed. Verify: The next scan fetches /.well-known/security.txt as text and the security.txt check passes.
Each file is written from what this scan found on the site. The content is in the JSON once the evidence is opened: https://www.seconddoor.io/api/reports/on-4592me
## The agent door
- **machine_discoverability** 96/100. No AI crawler is blocked by robots.txt; the site explicitly allows gptbot, claudebot, perplexitybot and others. A real browser is served the page normally. The sitemap lists 75 URLs including product pages.
- **programmatic_onboarding** 10/100. No API documentation is linked from the homepage. No OpenAPI spec exists at any of the 4 standard paths checked. The site publishes an MCP manifest but no API surface to call.
- **pricing_legibility** null/100. Could not verify: this scan did not reach the pages that would show it.
- **agent_aware_instrumentation** 58/100. An MCP manifest is published at /.well-known/mcp.json, signalling intent to work with agents. No Web Bot Auth signals are present, which is normal for a young convention.
- **machine_fetchable_trust** 40/100. No security.txt file is published; the path returns a web page instead. The homepage serves structured data to automated requests. Trust signals are minimal and incomplete.
- **commercial_rails** 10/100. No UCP manifest, no storefrontMCP, no payment processor detected on the homepage. No published route for an agent to complete a purchase. The site has no machine-readable checkout path.
## The human door
- **steps_to_first_value** null/100. Could not verify: this scan did not reach the pages that would show it.
- **required_fields** null/100. Could not verify: this scan did not reach the pages that would show it.
- **verification_walls** null/100. Could not verify: this scan did not reach the pages that would show it.
- **error_recovery** null/100. Could not verify: this scan did not reach the pages that would show it.
---
Scored by SecondDoor. Method: https://www.seconddoor.io/methodology
Scan another site: https://www.seconddoor.io
Rate an issue or this report: POST https://www.seconddoor.io/api/feedback, or the rate_fix tool on https://www.seconddoor.io/api/mcp