Is Agentic is a free tool from Vercel that checks how well AI agents can discover, read, and use your website.
Paste in a URL, and it returns a score from 0 to 100, along with the checks that passed, the problems it found, the evidence behind each result, and recommendations for fixing them.
The tool uses Ora AI for the underlying audit. Is Agentic then turns that evidence into its own score and organizes the report around the parts of the site that actually apply.
Open a failed check to see what the scan found and how to address it. After you select the issues you want to work on, Prompt to improve combines them into an implementation brief that you can take into a coding session, then rescan after the changes go live.
What Is Agentic Actually Checks
AI agents depend on many of the same web fundamentals as search crawlers and assistive software. They need a page they can reach, meaningful content in the HTML, useful document structure, real HTTP status codes, understandable controls, and clear information about what a site or product can do.
Is Agentic turns those requirements into three scoring groups:
| Group | Weight | What It Focuses On |
|---|---|---|
| Essential | 80 points | Access, readable content, HTTP behavior, navigation recovery, and usable controls |
| Recommended | 20 points | Relevant API, authentication, MCP, developer, commerce, and other agent-facing capabilities |
| Bonus signals | Up to 5 points | Newer agent-oriented formats and conventions |
Note that Not every check applies to every website. Is Agentic activates Recommended checks only after the scan finds evidence that the capability exists. A blog does not lose points for having no OAuth flow. A SaaS site with a public API can receive additional checks for documentation, schemas, errors, rate limits, or function-calling compatibility.
Open the individual findings to see why the site gained or lost points. Each failed or partial check shows the evidence found during the scan and, when available, a recommendation you can use as the starting point for a fix.
What You Get After a Scan
The report opens with the overall score and a short status such as Important blockers. Essential and Recommended results below it show how the score was calculated.
Essential failures deserve attention first. These checks look at basic conditions an agent needs before it can do much with the site. A blocked crawler, a JavaScript-only homepage, a misleading 404 response, or an unlabeled control can stop an agent long before API or MCP support becomes relevant.
Recommended findings become more specific to the site. If Is Agentic detects developer resources, the report can look for an OpenAPI description, machine-readable errors, rate-limit information, MCP discovery, API schemas, and other signals that help software understand how to interact with the product.
Each report can also include an observed agent task. The agent attempts a real navigation or information-gathering task and records where it encountered friction. That run does not affect the numeric score. It works as supporting evidence that you can compare with the technical checks.
The Prompt to improve option is especially useful when several issues need work. Select the findings you want to address and Is Agentic turns them into an implementation prompt that can be copied into a coding assistant or development task.
A Real Example: ScriptByAI.com Scored 54/100
A ScriptByAI.com scan run on August 23, 2026 returned a score of 54/100. The report gave the site 37/80 Essential points, 13.6/20 Recommended points, and +2.9 bonus points.
The scan flagged agent crawler reachability, 404 recovery, server-rendered content, OpenAPI discovery, and structured API errors. It also recognized several things that were already in place, including a sitemap, llms.txt, readable document structure, labeled controls, trust pages, and structured data.
Is Agentic also detected developer-related resources and a reachable GraphQL endpoint on ScriptByAI, which activated extra API checks for OpenAPI, JSON errors, schema documentation, and function-calling compatibility.
How to Use Is Agentic
1. Start at https://is-agentic.com/ and paste in the public URL you want to check. Click the Score button to start the scan. Ora AI performs the audit, then Is Agentic builds the report from the returned evidence.
2. Open the Essential section first and inspect any failed or partial checks. Compare the evidence with your actual site configuration before you change code, crawler rules, or server behavior.
3. Review the Recommended section and pay attention to the capabilities Is Agentic detected. API, GraphQL, OAuth, MCP, and developer-resource checks are most useful when those capabilities are part of the site you want agents to use.
4. When you are ready to make changes, select the relevant findings and copy the Prompt to improve prompt. Review any proposed code or configuration changes before deployment, especially changes to crawler rules, redirects, caching, authentication, APIs, or security controls.
5. After the update is live, run the scan again. The new report shows whether the original finding changed and can expose new issues introduced by the fix.
Using Is Agentic in Developer Automation
If you want the same report data inside scripts, CI tasks, coding agents, or other developer tools, Is Agentic exposes completed reports through several machine-readable formats.
| Access Method | What It Does | Typical Use |
|---|---|---|
| Browser | Creates and displays a report | Manual website review |
| CLI | Shows the report in a terminal and can start a scan when no report exists | Developer checks and scripts |
| Report API | Returns the latest completed report as JSON | Automation and integrations |
| MCP server | Exposes report, methodology, and developer-doc tools to MCP clients | Agent clients and automations |
| Markdown | Returns supported pages and reports as text/markdown | Text-oriented agents and crawlers |
CLI
The official npm package can retrieve a report directly from the terminal:
npx is-agentic example.comAdd --json when you need the structured response:
npx is-agentic example.com --jsonIf a completed report already exists, the CLI returns it immediately. When no stored report exists, the command starts a scan and waits for the completed result.
Report API
This API returns the latest completed report for a URL:
curl "https://is-agentic.com/api/v1/report?url=https%3A%2F%2Fexample.com"The API is read-only and does not start a new scan. It currently allows 120 requests per client IP in each 60-second window. Rate-limited requests return HTTP 429 with Retry-After information.
MCP Server
Any MCP client can connect to:
https://is-agentic.com/mcpThe MCP server exposes three read-only tools:
is_agentic_get_reportretrieves a completed report.is_agentic_get_methodologyreturns information about the score.is_agentic_get_developer_docsreturns integration guidance.
Agent Skill and Markdown
An official agent skill packages the Is Agentic guidance for compatible coding agents:
npx skills add vercel-labs/is-agenticSupported public pages and completed reports can also return Markdown when the request includes Accept: text/markdown:
curl -H "Accept: text/markdown" https://is-agentic.com/methodologyPublic Reports and Privacy
Is Agentic reports are public by design. When you submit a URL, the service sends it to Ora for analysis and stores the resulting scan information, which can include the target URL, score, findings, evidence, recommendations, timestamps, and report metadata.
Use the scanner only for URLs that are appropriate for public inspection. Avoid private application pages, authenticated URLs, confidential addresses, links containing secrets, and any URL whose technical findings should stay private.
What the Score Cannot Tell You
A scan reflects what was visible from one public URL at a particular time. Crawler rules, temporary outages, geographic behavior, site updates, external services, Ora’s checks, and Is Agentic’s scoring method can all change later.
Automated checks can also return false positives, false negatives, or partial results. The ScriptByAI example shows another edge case: capability detection can activate technically valid checks that need human judgment before they become implementation work.
The score is not a security audit, penetration test, accessibility certification, privacy review, legal-compliance assessment, or guarantee that every AI agent will complete every task on the site.
Alternatives & Related Resources
- Free AI Agent Readiness Checker & LLMs.txt Generator – IndexedAI
- Free Web-based MCP Server Inspector and Debugger – MCP Playground
- Is Your Site Agent-Ready? Free Tool to Audit AI Agent Compatibility
Pros
- Free to use with no account required for public scans.
- Turns technical findings into clear, prioritized fixes.
- Checks both basic web access and agent-specific signals.
- Supports browser, CLI, API, Markdown, and MCP workflows.
- Excludes checks that do not apply to the scanned site.
- Keeps evidence attached to each failed or partial check.
Cons
- Scan reports are public.
- Automated checks can produce false or partial results.
- Some detected APIs or endpoints may trigger irrelevant checks.
- One observed agent task cannot represent every AI agent.
- Scores can change when the audit methodology changes.
FAQs
Q: Does the Is Agentic API start a new website scan?
A: No. The report API reads the latest completed report. Start a new scan from the website, or use the CLI when no completed report exists for the requested domain.
Q: Why my score change after I rescan the same website?
A: The website itself may have changed, a time-sensitive check may return a different result, or the underlying audit and scoring methods may have been updated. Check the scan timestamp and individual findings before comparing two scores.
Q: Should I implement every recommendation in the report?
A: Review the evidence and the capability that activated each check first. A recommendation is worth implementing when it describes a public feature that you actually expect agents to discover or use.
Q: Does a low score mean a site is broken for human visitors too?
A: Not necessarily. The checks target what an AI agent needs, such as server-rendered HTML and machine-readable error pages, which sometimes overlaps with human usability and sometimes does not. The score itself is not an accessibility or quality certification.
