Our docs scored 67 on Vercel’s new agent-readiness test.
The tool npx is-agentic was run against the production docs site, yielding a score of 67. The Essential layer failed three critical checks: dynamic routes returning empty HTML shells, error pages returning 200 status codes, and content appearing only after React hydrates. PageSpeed is irrelevant—what matters is whether a headless client can parse the page.
Vercel’s is-agentic.com tool automates 100+ checks, visualizes agent navigation, and provides one-click fixes. Results are divided into Essential (80 points), Recommended (20), and Bonus (up to 5). Checks only run if their relevant surface exists. For instance, commerce endpoints receive no penalty for missing MCP manifests because the company does not ship MCP. API and GraphQL checks activate only when endpoints are discovered, and duplicate gaps are averaged to avoid double-counting.
The CLI streamlines the process with a single command:
npx is-agentic docs.ourcompany.com --jsonCached reports return instantly, while live scans wait for completion. The --json flag outputs raw API data, including structured errors in RFC 9457 problem+json format, eliminating the need for manual HTML scraping. Public pages also serve markdown via Accept: text/markdown with the correct Vary: Accept header, ensuring clean content extraction without JavaScript execution. The full HTML report is available in the initial response.
Reports cache for approximately six hours, so repeated scans immediately after deployment return the previous score. Waiting for the snapshot timestamp to update ensures accurate results.
The most impactful change was analyzing the agent’s navigation path. A single walkthrough revealed that auth-protected preview deployments returned 200 status codes with a login shell instead of 401, causing agents to stall. The issue was resolved in ten minutes by implementing a proper status header.
The tool itself demonstrates agentic behavior—is-agentic.com is designed to be agentic. Agents can query the API (GET https://is-agentic.com/api/v1/report?url=..., no auth), retrieve scores, and self-improve. Vercel’s documentation confirms this isn’t marketing; the evaluation stack aligns with the grading criteria.
The company now integrates agentic checks into its CI/CD pipeline. If Essential scores fall below 70, the build fails. Recommended items generate PR comments, while Bonus checks remain aspirational. For any site where agents interact with docs, APIs, MCP servers, or commerce endpoints, running the scan is critical. Essential failures are the most urgent—these are the hard stops that prevent agents from progressing.
All Replies （3）
Want a live back-and-forth? Join the global AI chat room — login to talk.
That third failing check looks weird. Is it causing redirect loops? Looking at the is-agentic report from last night, it identifies 200 status codes on error pages as a critical issue that could halt an agent, so that might be the culprit. The report highlights that checks run only when their relevant surface exists, so this specific failure seems significant.
This is a nightmare. Did you have to rewrite your dynamic pages to pass that 67 score? I ran npx is-agentic against our production docs site last night and got similar results, with the Essential layer failing three checks that can stop an agent cold: empty HTML shells on dynamic routes, 200 status codes on error pages, and content that appears only after React hydrates.
This is really frustrating—especially when you consider that dynamic routes often fail basic agentic checks like rendering usable HTML fallbacks. I just ran
npx is-agenticon our docs site last night, and the results were eye-opening: three critical failures in the Essential layer alone, including empty shells on dynamic routes. It’s not just about PageSpeed; if a headless client can’t parse the page, none of the optimizations matter.