Vercel’s Is Agentic Audits Website Readiness for AI Agents

Vercel’s Is Agentic is a free public scanner that estimates how easily AI agents can discover, access, understand, and use a website. The tool is powered by Ora’s agent-experience research and turns the results into a score, evidence-backed findings, recommendations, and an observed agent journey. For an operator, the practical use is straightforward: scan a public URL, fix the highest-value friction, and rescan rather than treating a single number as a certification. MarkTechPost’s launch coverage reports the release details, while the official Is Agentic site exposes the public scanner and reports.

Definition: Is Agentic is Vercel’s free website audit for AI-agent discoverability, access, understanding, and use.

Example: A company can scan its public site and receive findings about server-rendered content, HTTP behavior, structure, controls, and interfaces that agents can use.

Key takeaway: The score is a prioritized readiness snapshot, not a universal quality badge.

Business impact: Teams can use the report to decide which public-site fixes should come before an agent, API, MCP, or commerce rollout.

What does Is Agentic measure?

Is Agentic measures the public surfaces an AI agent can reach and operate, with essential checks covering server-rendered content, HTTP behavior, document structure, recoverable errors, and usable controls. Recommended checks activate only when scan evidence identifies a capability such as an API, OAuth flow, GraphQL endpoint, MCP server, developer portal, or commerce surface. That applicability rule matters for business sites because Is Agentic does not count an interface as a failure when the site does not claim to provide it.

The tool also separates the numeric score from an observed agent journey. Is Agentic displays where one agent encountered friction, but the official methodology says that one task cannot represent every agent, user goal, or production condition. The takeaway is to read the evidence and journey together: the score helps prioritize, while the journey shows one concrete path that still needs investigation.

How many checks does Ora run?

Ora’s methodology describes 118 checks across four layers: Discovery has 15 checks worth 20 points, Access has 41 checks worth 30 points, Usability has 56 checks worth 40 points, and Payments has 6 checks worth 10 points. The Ora methodology says the checklist is reverse-engineered from observed agent behavior across many intents and sites, rather than written as an opinion-only checklist.

Vercel’s Is Agentic uses a different presentation of those results. Essential checks share an 80-point pool, Recommended checks share a 20-point pool, and Emerging signals can add a bonus capped at five points without becoming requirements. Is Agentic also excludes not-applicable checks, gives partial results proportional credit, and averages duplicated check IDs across MCP surfaces. Operators should therefore compare the evidence and score breakdown, not assume that a raw pass count maps directly to the headline score.

What does an Is Agentic report contain?

An Is Agentic report combines a score with individual findings, observed evidence, and recommendations where available. The public report format can expose a score label, scan timestamp, eligible checks, score breakdown, and issue records that identify the check, result, tier, details, and recommendation. That structure makes the report more useful than a badge: an engineering team can turn a failed check into a specific remediation task.

The report also supports machine consumption. Completed reports can be retrieved through public JSON and Markdown representations, and the Is Agentic MCP server exposes read-only report, methodology, and developer-documentation tools. The official product description says these interfaces do not start scans or mutate the target site. A team can use them to feed report evidence into internal review workflows without granting the scanner credentials to change the audited website.

Is Is Agentic free for teams?

Is Agentic is currently presented as free, with browser scanning and public report access available without a billing account or API key. That lowers the barrier for a startup that wants a baseline audit, while the machine interfaces give larger teams a path to repeat checks around documentation portals, developer sites, APIs, or commerce surfaces.

Free access does not remove the need for engineering judgment. A score describes evidence from a public URL at a particular point in time; status codes, rendered content, crawler policies, documentation, APIs, and agent protocols can change later. Teams should record the scan timestamp, fix the underlying issue, and rescan the same URL to check whether the public behavior actually improved.

What should a business do with the score?

A business should start with failed Essential checks, then inspect Recommended checks that match the capabilities its website actually advertises. This sequence follows Is Agentic’s own guidance and avoids spending time on an API, OAuth, MCP, GraphQL, or commerce requirement that the site does not offer.

The score should sit inside a broader launch checklist rather than replace it. Is Agentic explicitly does not certify security, accessibility, privacy, legal compliance, business quality, or compatibility, and the tool cannot observe private pages, authenticated workflows, geographic differences, temporary outages, or user-specific state. Use the audit to improve the public agent-facing layer, then validate the real workflow with dedicated technical and human review. For background on the systems agents connect to, see what an AI agent is and how MCP connects AI applications to tools and data.

What changes for website operators?

Is Agentic makes agent-readiness visible as an operational web-quality problem. A site that renders useful content only after JavaScript, returns misleading error statuses, hides its structure, or exposes an interface agents cannot understand may still look fine to a human visitor while failing an automated task. A report gives operators a concrete starting point for closing that gap.

The most defensible interpretation is not “a high score means agents will always succeed.” It is “the public evidence currently shows fewer of the measured failure modes.” That distinction keeps the tool useful: teams can prioritize fixes, compare fresh scans, and test their own workflows without turning a changing research score into a promise of autonomous success.