Vercel launches free tool to assess website readiness for AI agents

Vercel (Vercel) released Is Agentic, a public-facing tool that assigns a numeric score indicating how prepared a site is for discovery, access, understanding and use by AI agents. The scanning and scoring are performed by Ora, a research firm specializing in agent experience from era labs; Vercel provides the interface, result reports, storage and the logic that produces the final rating. The tool is available at no cost, without paid plans, without an API key and without a billing account; users can enter a URL in a browser or run it from the CLI with `npx is-agentic <domain>` and receive JSON output using the `--json` flag.

Four layers and 118-plus tests

Ora’s methodology splits an agent’s journey into four layers: Discovery (20 points across 15 tests), Access (30 points across 41 tests), Usability (40 points across 56 tests) and Payments (10 points across 6 tests). The total of 118 tests aligns with Vercel’s claim of “100+ tests”. Ora states the list is derived from real-world agent runs rather than subjective opinion. Letter grades are assigned as follows: A+ 95-100, A 86-94, B 70-85, C 48-69, D 28-47, F 0-27.

Vercel’s scoring model

Vercel reorganizes the tests into its own display model: Essential tests share an 80-point pool, Recommended tests account for 20 points, and Emerging signals add a bonus of up to 5 points when present but never subtract points when absent. Non-relevant tests are excluded from the calculation instead of being counted as failures, and partial results receive proportional credit. This logic matters: Recommended tests are only activated when the scan confidently detects an API, an OAuth flow, a GraphQL endpoint, an MCP server, a developer portal or a commerce surface; a marketing site is not penalized for lacking an interface it never declared.

What the report contains

Each finding is accompanied by observed evidence and, where applicable, a concrete remediation suggestion. A live API response reveals the structure: score, label, scan time, number of eligible tests, point breakdown by tier, and an array of issues. Every issue carries an identifier, name, description, recommendation, result and grade. Real test identifiers include `content-no-js`, `agent-friendly-404`, `markdown-negotiation-vary`, `json-ld`, `sitemap`, `trust-anchors`, `metadata-completeness`. The 404 test requires an actual HTTP 404 response, not a 200 that serves an application shell. Reports also embed the actual agent journey observed, displayed separately from the numeric score because a single task does not represent every agent, and a “Prompt to fix” mechanism turns practical recommendations into implementation briefs for code agents.

Machine interfaces and limits

Three read-only endpoints expose completed reports: the API at `/api/v1/report` with a rate limit of 120 requests per IP address in a 60-second window, error responses conforming to RFC 9457 with stable codes such as `invalid_url` and `report_not_found`, as well as the MCP server and CLI. The tool targets any organization that runs a public web surface, from seed-stage startups receiving a baseline audit without purchase, through SaaS teams that pipe JSON output into CI as a regression gate, to enterprises benchmarking documentation portals, developer sites and commerce surfaces across business units. Use cases include pre-launch audits, CI checks that fail when server-rendered content regresses, competitive benchmarking against public scores, verification that an MCP or OpenAPI surface is discoverable, and prioritizing technical fixes with ready-to-copy prompts.