AI API Integration Gets a New Agent-Readiness Score
Vercel and Ora launched Is Agentic on August 23, 2026, giving teams a free way to assess AI API integration readiness across public websites. The release matters because agent discovery and usability issues increasingly affect how products are found, cited, and completed by machine actors, not just human visitors. According to MarkTechPost’s coverage of the launch, the tool is free to use across the web interface, CLI, API, and MCP server.
Vercel launches Is Agentic for public websites
The product is straightforward: enter a domain, or run npx is-agentic <domain>, and the system returns a score showing how well AI agents can discover, access, understand, and use that site. Vercel operates the interface and report delivery, while Ora provides the scoring methodology and audit logic.
That distinction matters for product and engineering teams. This is not another generic site health score. It is aimed at agent-facing paths such as documentation portals, commerce flows, APIs, and machine-readable content surfaces. For teams thinking about AI integration services or preparing AI API-first interfaces, it offers a baseline that is easier to test than to debate.
The source article also notes an important caveat: Vercel says the score is “not a certification, security audit, or accessibility review.” In practice, that positions the output as a technical prioritization tool rather than a pass-fail badge.
How Ora’s 118 checks map to agent-readiness
Ora’s model spans four layers that mirror the way agents move through a site. Discovery accounts for 20 points across 15 checks. Access covers 30 points across 41 checks. Usability carries the largest weight at 40 points across 56 checks. Payments adds 10 points across 6 checks.
That is 118 checks in total, which explains Vercel’s broader claim of “100+ checks.” Ora also publishes a letter-grade scale: A+ for 95 to 100, A for 86 to 94, B for 70 to 85, C for 48 to 69, D for 28 to 47, and F for 0 to 27. The article says the checklist is reverse-engineered from real agent runs rather than assembled as a theoretical best-practice list.
For implementation teams, that is the more interesting point. Many AI connectors fail for reasons that do not show up in classic SEO or frontend QA: JavaScript-only content, poor status handling, weak metadata, absent sitemaps, or machine interfaces that exist but are hard for agents to identify. Those gaps sit squarely in AI integration architecture, especially when a company expects copilots, shopping agents, or research agents to interact with public assets reliably.
Which teams should run the audit first
The tool appears broadly deployable because it is free and unauthenticated. Startups can use it as a baseline check before launch. Mid-market SaaS companies can move the JSON output into CI and fail builds when agent-facing regressions appear. Larger enterprises can compare public surfaces across brands, business units, or product lines.
The immediate fit is strongest in developer tools and SaaS, e-commerce and retail, and travel and hospitality. Those sectors depend heavily on structured content, clear navigation, and transactional paths that an agent may try to search, compare, book, or cite. In each case, enterprise AI integrations increasingly start with the public surface before they expand into account-level workflows.
There is also a practical operational insight here: a low score on a public site can point to downstream friction in product-led acquisition. If an agent cannot parse a pricing page, documentation set, or booking flow, the issue is not just visibility. It can become a conversion and support problem. That is where AI implementation services start to look less like a future roadmap item and more like routine web operations.
Why Vercel’s scoring model matters for implementation teams
Vercel does not simply display Ora’s raw weighting. The report groups findings into Essential checks worth 80 points, Recommended checks worth 20 points, and Emerging signals that can add up to 5 bonus points. It also excludes not-applicable checks rather than treating them as failures.
That design avoids one of the biggest objections teams usually have to external scoring systems: unfair penalties. A marketing site is not marked down for lacking a commerce flow or an API it never claimed to expose. Recommended checks only activate when the scan finds evidence of an API, OAuth pattern, GraphQL endpoint, MCP server, developer portal, or commerce surface.
For teams responsible for AI deployment services, this applicability logic is more than a user-experience detail. It changes remediation order. Essential checks indicate broad agent friction, while Recommended checks point to deeper fixes on technical surfaces the organization already exposes. Partial credit also matters, because it reflects the reality of rollout work: a site may be moving toward better machine access without being fully there yet.
This is where implementation discipline matters more than the score itself. A report that flags content-no-js, agent-friendly-404, json-ld, or metadata-completeness is often pointing at different owners across content, frontend, platform, and API teams. Without a clear workflow, the score can become another dashboard number that does not produce change.
What the reports reveal about machine interfaces
The report surfaces are unusually practical. According to the source article, teams can access completed reports via HTML pages, Markdown responses, JSON from the read-only API, CLI output, and an MCP server. The report API is rate-limited to 120 requests per client IP per 60-second window and uses RFC 9457 problem details for structured errors.
Vercel also points to OpenAPI as the supported integration contract, while API discovery is exposed through RFC 9727 and deprecation signals follow RFC 9745. Those standards details may sound narrow, but they matter for teams wiring reports into developer workflows and long-term monitoring.
A notable feature is the inclusion of observed evidence and a recommendation for each issue. Reports can include score, score label, scanned timestamp, eligible checks, breakdown by tier, and issue-level fields such as name, recommendation, result, and details. The article also notes a “Prompt to fix” option that turns findings into an implementation brief for a coding agent.
That is the strongest operator signal in the launch. The real value is not that public websites now have one more score. It is that teams can move from detection to remediation in a format machines can also consume. For organizations already working on custom AI integration tailored to their business, that shortens the gap between finding friction and fixing it across websites, APIs, and agent-facing flows.
What this means for implementation teams next
The most useful way to read this launch is as an early operations tool for agent-facing surfaces. It gives teams a common score, but more importantly a shared issue list across content, web, and API owners. That can help resolve a recurring problem in AI API integration work: nobody sees the full path from public discovery to successful machine use.
The trade-off is that a readiness score is still only a proxy. It cannot represent every agent task, every session path, or every commercial edge case. But if teams treat it as a backlog generator rather than a badge, it can become a practical part of release management.
What to watch next is whether teams start treating agent-readiness checks like performance and accessibility gates in CI, and whether the broader ecosystem standardizes around similar evidence formats. If that happens, public web surfaces may become a more formal part of AI integration architecture reviews, not just marketing-site maintenance.
Related reads
Martin Kuvandzhiev
CEO and Founder of Encorp.io with expertise in AI and business transformation