The score
What is agent readiness?
Agent readiness is whether an AI agent can find, read and use a site without a person. The StartupHub.ai Agent Readiness scanner is free and needs no signup. Paste a website, REST API or MCP endpoint and it returns a score from 0 to 100 with a letter grade. It runs 44 checks in six categories and every failed or warned check comes with a fix prompt for Claude Code or Cursor.
StartupHub.ai has run 113,581 scans covering 93,652 sites. Of the 4,328 sites scanned in the last 45 days the median score was 50 out of 100, 79% graded D or F and 1.8% reached A or A+.
What it checks
The six agent readiness categories
The scan covers six categories. A passed check earns its full weight, a warning earns 40 percent, a failed check earns nothing and a skipped check is left out. The weighted total is 0 to 100: A+ is 90 or above, A is 80, B is 70, C is 60, D is 40 and F is below 40.
Discoverability
This category checks whether an agent can discover the site. It verifies a robots.txt file, a declared sitemap, a Link response header that follows RFC 8288, a rel="describedby" link to llms.txt and JSON-LD SearchAction for the on-site search URL. AGENTS.md or CLAUDE.md counts as an optional pass for code repositories.
Content
This checks if agents can read content. It verifies markdown for Accept: text/markdown, llms.txt in v2 shape with H1, summary and links with a short note each, that those links open, optional llms-full.txt, that page text is in raw HTML, that headline and first facts sit in first 50 KB and that HTML stays under about 300 KB because many AI fetchers stop after a few hundred KB.
Access control
This category checks whether AI agents can access the site. It verifies that Claude-User, Claude-SearchBot, ChatGPT-User and OAI-SearchBot are not blocked at the site root, that robots.txt has explicit rules for AI crawlers and that Content Signals are declared. It also fetches the page as GPTBot, ClaudeBot, PerplexityBot, meta-externalagent, Bingbot and Googlebot to confirm each one succeeds, while training crawlers can stay blocked.
Capabilities
This category checks what an agent can do beyond reading. It verifies an MCP server card, an API catalog that follows RFC 9727, an OpenAPI document, an Agent Skills index, OAuth discovery and agentic registration so an agent can get scoped access without a person provisioning keys. An A2A agent card is optional and never lowers the score.
Commerce (agent payments)
This category checks whether an agent can pay. It verifies x402, MPP, AP2 or ACP discovery, x402-mesh, Offer or Product JSON-LD with a price, a payment link reachable from the homepage and whether /checkout or /pricing exists. Each protocol is probed on its own and a site with no paid endpoints is not penalised for MPP.
Quality
This category checks quality signals that help an answer engine quote the page. It verifies complete Open Graph tags, a single H1, image alt text, a meta description, a canonical URL, JSON-LD, entity JSON-LD in the first 30 KB, RateLimit headers that follow RFC 9331 and security.txt.
Worked example
Agent readiness report example: stripe.com scores 57 out of 100
The scanner checked stripe.com on 22 October 2026. It scored 57 out of 100, grade D, and you can re-run the same check in the scanner above.
| Category | Score | Checks passed |
|---|---|---|
| Discoverability | 76 / 100 | 2 of 5 |
| Content | 69 / 100 | 5 of 8 |
| Access control | 76 / 100 | 2 of 4 |
| Capabilities | 27 / 100 | 0 of 6 |
| Commerce | 40 / 100 | 1 of 6 |
| Quality | 69 / 100 | 6 of 9 |
| Overall | 57 / 100 | Grade D |
Of 44 checks, 16 passed, 17 warned, 5 failed and 6 did not apply. The five failures were no markdown on Accept: text/markdown, no MCP server card, no Offer or Product JSON-LD on the homepage, two H1 tags, and 27 of 31 images without alt text. The homepage HTML is 670 KB and costs about 170,897 tokens to read as HTML. The markdown version would cost about 3,926, which is 98% fewer. stripe.com passes the llms.txt checks, with 119 linked files.
What the Live Agent Run found on stripe.com
| Agent | Answers found | Fetches | Time |
|---|---|---|---|
| Cursor | 8 of 8 | 8 | 95 s |
| Claude Code | 6 of 8 | 8 | 79 s |
| OpenCode | 6 of 8 | 7 | 26 s |
| Codex CLI | 5 of 8 | 8 | 75 s |
The Live Agent Run on stripe.com on 14 October 2026 asked 8 questions, because the scan found a way to pay. Cursor found all 8 answers. Claude Code and OpenCode each got 6. Codex CLI got 5. The plain model with a fetch tool was refused or got no usable text in four of its five attempts. So a site can grade D on the scan while capable agents still find most of what they need, and a simple fetcher gets turned away.
Fix first
How to improve your agent readiness score
- 1Publish an MCP server card. This is the heaviest single check at weight 5. If you run an MCP server, publish /.well-known/mcp/server-card.json describing its tools so agents such as Claude and Cursor can discover them.
- 2Serve markdown to agents. Your server should return the page as markdown when a request sends Accept: text/markdown. The check carries weight 4 and on stripe.com it would cut the homepage from about 170,897 tokens to about 3,926.
- 3Publish llms.txt in the v2 shape. Publish /llms.txt in the v2 shape with an H1 that includes the site name, a one-line summary and links that each carry a short note, and point at it with a rel="describedby" link. Doing both passes all four llms.txt checks for the guide, link notes, links resolve and discovery link.
- 4Put your prices in JSON-LD. This check has weight 3. Put JSON-LD Offer or Product data with price and priceCurrency on the homepage so an agent can read your pricing without scraping the layout.
- 5Keep the HTML small and the facts early. Keep HTML under about 300 KB and put the H1 and first facts in the first 50 KB. Many AI fetchers stop reading after a few hundred KB, so a footer, FAQ or contact details near the bottom of a heavy page may never be read.
Compared
StartupHub.ai scanner vs Cloudflare's isitagentready.com
Cloudflare runs a free agent readiness checker at isitagentready.com. Its page lists five categories: Discoverability, Content Accessibility, Bot Access Control, Protocol Discovery and Commerce. The StartupHub.ai scanner runs 44 checks in six categories, adding Quality: Open Graph, a single H1, image alt text, JSON-LD near the top and the size of the HTML. It does two things the isitagentready.com page does not list: the Live Agent Run with six agent frameworks and fetching the page as 17 named AI clients. It adds a public leaderboard, REST API and MCP tool, and both checkers are free. Full comparison.
Questions
Frequently asked questions
What is an agent readiness scanner?
+
An agent readiness scanner measures how well agents can use your site. The StartupHub.ai Agent Readiness scanner is free with no signup and scores a website, REST API or MCP endpoint from 0 to 100 with a letter grade. It runs 44 checks in six categories and includes a fix prompt for Claude Code or Cursor. It is also available as POST /api/v1/agent-readiness and as the MCP tool scan_agent_readiness.
Is the agent readiness scanner free?
+
Yes, the scanner is free with no signup. Paste a website, REST API or MCP endpoint and get a score from 0 to 100 with a letter grade. Live Agent Run is three a month free and Pro Mini from $5 a month removes the cap. POST /api/v1/agent-readiness costs 1 credit and the free plan includes 1,500 credits a month, and agents with no key can pay $0.001 over x402.
What is the Live Agent Run?
+
The Live Agent Run points six agent frameworks at your site from StartupHub.ai hardware. The frameworks are Claude Code, Codex CLI, OpenCode, Cursor, Gemini CLI and a plain model with a fetch tool. They fetch pages, read llms.txt, MCP server card and API docs and call what they find with nothing simulated. Each framework answers seven questions and an eighth about payment when found. The same run checks 17 named AI clients and the Live Agent Run shows what agents actually find.
Which AI agents run in the Live Agent Run?
+
The Live Agent Run uses six agent frameworks. They are Claude Code from Anthropic, Codex CLI from OpenAI, OpenCode open source, Cursor, Gemini CLI from Google and a plain model with a fetch tool as a control. Cursor runs on its own backend and the other frameworks run on models StartupHub.ai chooses. Windsurf and Cline are not included because they only run inside an editor.
Why is a plain model part of the Live Agent Run?
+
The plain model with one fetch tool is the Live Agent Run control. It is what teams build to point an AI at a site. If it finds as much as Claude Code on a site, the framework was not the variable and the site was. Across 1,307 runs since 1 October 2026 it averaged 2.9 answers on about one fetch while the agent frameworks found between 4.7 and 5.2 on five to nine fetches.
How is the agent readiness score calculated?
+
The agent readiness score is a weighted total from 0 to 100 across 44 checks in six categories: discoverability, content, access control, capabilities, commerce and quality. A passed check earns full weight and a warning earns 40 percent of its weight. A failed check earns nothing and a skipped check is left out of the total. Grades are A+ 90 or above, A 80, B 70, C 60, D 40 and F below 40. A link to startuphub.ai in llms.txt adds 8 points and a missing link never lowers score.
Does my site need an MCP server to score well?
+
No. MCP is part of one category, capabilities, out of six. A content site with good discoverability, content, access and quality scores well without one. An MCP server card matters when you have an API or a product an agent would want to call. It is the heaviest single check, weight 5.
How is agent readiness different from SEO?
+
They are different goals. SEO is about ranking in search results that people click. Agent readiness is about whether an AI agent can find, read and use a site without a person. That includes markdown, llms.txt, MCP and OpenAPI, machine-readable prices, and fetchers that do not run JavaScript. They overlap on JSON-LD, sitemaps, Open Graph and a single H1, which both use.
Can I run the scanner from code or an AI agent?
+
Yes. POST /api/v1/agent-readiness takes a URL and returns the score, grade, six category scores and every check as JSON. It costs 1 credit with a Bearer key, and the free plan includes 1,500 credits a month. Agents with no key can pay $0.001 a scan over x402, and the same scan is the MCP tool scan_agent_readiness on the StartupHub.ai MCP server. Live Agent Run results for a domain are readable at GET /api/agent-readiness/agent-task?domain=, one row per framework.
Does the scanner check x402 and MPP payments?
+
Yes, it checks whether an agent can pay. It checks x402, MPP, AP2 or ACP discovery, x402-mesh, Offer or Product JSON-LD with a price, a homepage payment link, and /checkout or /pricing. A site with no paid endpoints is not penalised for MPP, and each protocol is probed on its own. MPP is the Machine Payments Protocol from Tempo and Stripe, detected by a WWW-Authenticate: Payment challenge or /.well-known/payment.
Where can I see how other sites score?
+
You can compare on the public leaderboard on this page. It ranks scanned sites by overall score, by each of the six categories, and by Live Agent Run results. A site needs runs from at least three frameworks to appear on the Live Agent Run board. Each domain has a shareable report at /agent-readiness?url= followed by the domain.
Do you keep the sites I scan?
+
Yes, every scan is stored with its result. The latest score for a domain can appear on the public leaderboard, and on its StartupHub.ai profile when it has one. Only public pages are fetched, and the scanner refuses private network addresses.
