Static pre-check · Fri, 07 Aug 2026 06:49:42 GMT
AI agents struggle discovering and understanding site content due to missing meta descriptions and no llms.txt file guiding agent interactions.
Watch a real agent use your site →
A live Claude Code run against chroma.com, streamed step by step.
Measure this with real agents →
An audit: measured success rates from repeated real agent runs, with confidence intervals and downloadable evidence.
no <meta name="description">
Fix: Add a <meta name="description"> of 50–200 characters summarising what the product does. Agents use it as the one-line answer to "what is this site".
no link to an API reference in the homepage HTML
Fix: If you have an API, link its reference from the homepage. If you don't, this check is safe to ignore.
title is 96 characters (truncated in most surfaces)
Shows the current scan score and grade for this domain, straight from the latest static pre-check.
[](https://www.vorza.dev/score/chroma.com)<a href="https://www.vorza.dev/score/chroma.com"><img src="https://www.vorza.dev/api/badge/chroma.com" alt="agent readiness score" /></a>Fix: Trim the <title> to 70 characters or fewer so it is not cut off.
https://www.chroma.com/
2 docs link(s), e.g. /resources/technical-library/guides-and-instructions/cleaning-filters/
og:title and og:description present
all 6 agent crawlers may fetch /
HTTP 200, 893 bytes
robots.txt has a Sitemap: line
7 URL(s) listed
search-based checks are deferred — excluded from the score
registry lookup not implemented in v1 — excluded from the score rather than guessed
/llms.txt not served (HTTP 404)
Fix: Publish /llms.txt: a plain-markdown index that tells an agent what this product is and where the authoritative docs live. It is the cheapest high-leverage file you can add.
no JSON-LD blocks on the homepage
Fix: Add a JSON-LD block (schema.org Organization or SoftwareApplication) so agents get your name, description and links as data rather than prose.
/llms-full.txt not served
Fix: Optional but cheap: publish /llms-full.txt with your docs inlined as one markdown file, so an agent can read everything in a single fetch.
Accept: text/markdown → text/html; charset=UTF-8
Fix: Optional but increasingly expected: honour `Accept: text/markdown` by serving a markdown rendering of the page, so agents skip HTML parsing entirely.
ratio 0.0392 — 9688 readable chars in 246957 bytes of HTML
Fix: Most of what you serve is markup and inline script rather than content. Reduce inline payloads or server-render more text so the signal-to-noise ratio favours readers.
HTTP 200 for / (agent UA)
https://www.chroma.com/
both User-Agents got HTTP 200
9688 characters of text in the served HTML
1 redirect(s)
/docs not served (HTTP 404) — docs may live on another path or host
no llms.txt
no llms.txt to parse
/agents.md not served (HTTP 404)
Fix: Publish /agents.md: what an agent may do here, which endpoints matter, auth requirements, and rate limits. It is the agent-facing counterpart to a README.
/docs → HTTP 404
Fix: Serve docs at the conventional /docs path, or redirect it to wherever they live. Agents try conventional paths first.
404 returns HTML, not a machine-readable body
Fix: Content-negotiate your errors: return a JSON body with a stable `code` (and a `next_action` where recovery exists) to clients that ask for JSON.
/openapi.json not served (HTTP 404)
Fix: If you have an HTTP API, publish its OpenAPI document at a predictable path (/openapi.json). It is the difference between an agent reading your contract and an agent guessing it. If you have no HTTP API, ignore this.
/.well-known/mcp.json not served (HTTP 404)
Fix: If you run an MCP server, advertise it at /.well-known/mcp.json so agents discover it without being told its URL.
/api → HTTP 404 (no API at the conventional root)
no MCP endpoint found to probe (none in the input URL and none advertised at /.well-known/mcp.json)
no listed MCP tools to inspect
no listed MCP tools to inspect
no MCP endpoint to probe
no parseable OpenAPI document
no parseable OpenAPI document
no parseable OpenAPI document
no parseable OpenAPI document
no agent-payment rail published — not applicable. This layer is excluded from the score rather than counted against the site.
no agent-payment rail published — not applicable. This layer is excluded from the score rather than counted against the site.
no agent-payment rail published — not applicable. This layer is excluded from the score rather than counted against the site.