Agent-ready website: why AI agents will soon be your most important visitors – and how they can send you a request directly

What is an agent-ready website?

An agent-ready website serves software agents as well as humans: it provides content in machine-readable form, tells agents what they are allowed to do, and offers a defined interface through which an agent can perform actions – for example send a consultation request. The difference from classic SEO: SEO makes you findable. Agent readiness makes you understandable, quotable and actionable for an agent.

The term is not a marketing invention. It rests on concrete open standards that became established in 2025/2026: llms.txt as a short profile for language models, Markdown delivery via content negotiation, Anthropic’s Model Context Protocol (MCP) as the interface standard, Agent Skills, the Agent Registry Descriptor (ARD) and auth.md for agent registration. Scanners such as isitagentready.com test websites against exactly these standards and assign levels from 1 (browser only) to 5 (Agent-Native).

• Audience 1: humans with a browser – the classic website
• Audience 2: AI agents – need plain text, structure and interfaces
• SEO = being found; agent readiness = being understood and usable
• Based on open standards, not vendor tricks

Why is normal SEO no longer enough for AI agents?

An agent works differently from a human. It has no eye for your design, no patience for cookie banners and no cursor for dropdown menus. It receives your page as text, often truncated, and must decide in seconds: is this provider relevant, what does it cost, how do I get in touch? Anything hidden in images, JavaScript widgets or behind clicks does not exist for it.

Then comes the task at the end of the research. More and more agents are expected to act, not just inform: “Request a quote for a practice website from three web design providers in Berlin.” A classic contact form is a barrier for agents – inconsistent field names, captchas, mandatory fields without explanation. Without a defined interface the agent gives up or picks the competitor who makes the request easy. This is where the competitive advantage for early-moving small businesses emerges over the next few years.

• Agents see text and structure, not visuals
• Cookie banners, captchas and widgets are obstacles
• Agents increasingly act, not just read
• Whoever makes the request easy gets it

Which building blocks does an agent-ready website need?

The good news: it is a manageable number of files and endpoints, and all of them can be generated from content your website already has. The principle is “one source of truth”: prices, services, contact details and FAQ are maintained once and delivered both to the HTML pages and to the agent formats. Otherwise, after the next price update your website is correct but the agent reads old numbers.

The key building blocks, from simple to advanced:

• robots.txt with clear rules for AI crawlers (e.g. GPTBot, OAI-SearchBot, ClaudeBot) plus a Content-Signal line: search yes, AI input yes, training no
• llms.txt: a text file in the root with company profile, services, prices, contact and a list of key pages – the short profile for language models
• Markdown for Agents: every page is delivered on request (Accept: text/markdown) as clean plain text without navigation, footer and scripts
• Structured data (JSON-LD): LocalBusiness, Service, FAQPage, Article – so facts are unambiguous
• API catalog (/.well-known/api-catalog) and Agent Registry Descriptor (ARD): the table of contents of all agent resources
• Agent Skills (SKILL.md): instructions on how an agent should answer about your business – including notes such as “emergencies: have them call” or “prices exclude VAT”
• MCP server: the actual interface through which agents fetch data and perform actions
• auth.md and OAuth: how an agent registers and authenticates so requests remain traceable and limited

How can an AI agent send a request directly through an interface?

This is the part that surprises most business owners: an agent does not have to fill in your contact form. It can leave a request directly via a defined interface – with name, callback number or e-mail, concern and consent. The standard for this is the Model Context Protocol (MCP). An MCP server on your domain exposes so-called tools, for example get_plans (fetch packages and prices), search_faq (search frequent questions), get_article (read a guide in full) and request_consultation (send a consultation request).

In everyday use it looks like this: a customer tells their assistant “Find me a provider for a website subscription and request an initial consultation.” The agent finds your domain, reads llms.txt, checks prices via get_plans, clarifies open questions via search_faq – and then calls request_consultation with the customer’s contact details. A normal e-mail lands in your inbox, flagged as an agent request, with all the details. You call back as with any other enquiry. The difference: the customer never opened your website.

To stop this becoming a spam channel, registration is part of the design. Via auth.md the agent learns how to sign up: anonymous registration with an agent name, a short-lived access token, clearly limited rights (read plus send request). Technical limits apply on top: a few registrations per hour and IP, at most three requests per registration, mandatory consent, validation of phone or e-mail, at most two links in the text. Every request can be traced to a registration and revoked. The channel stays usable – and GDPR-compliant, because only the data the customer deliberately gave to their agent is transmitted.

• MCP server = toolbox for agents on your domain
• Read tools: company info, plans, FAQ, page list, articles
• Action tool: send consultation request – arrives as e-mail
• Registration via auth.md, tokens with limited rights
• Rate limits, mandatory consent and revocation against abuse

How do you check whether your website is agent-ready?

You do not need a developer for this. Three checks give a first impression. First: open your-domain.com/llms.txt. If a readable text file with company profile and prices appears, the first building block is there. If you get a 404, language models only know what they happen to extract from your HTML. Second: open your-domain.com/.well-known/mcp/server-card.json – if there is a description with tools, an agent can interact with you. Third: use a scanner such as isitagentready.com, which tests all standards automatically and assigns a level from 1 to 5.

For orientation: most German SME websites today sit at level 1 to 2 – crawlable, nothing more. Level 3 to 4 is reached by implementing llms.txt, Markdown delivery and structured data cleanly. Level 5 “Agent-Native” requires a working MCP server, Agent Skills, ARD and a registration option. Our own website bezahlbare-webseite.de has been verified at level 5 since autumn 2026 – with six MCP tools including consultation request. The reference bestattungen-in.berlin and further client projects follow the same architecture.

• Test 1: open /llms.txt – readable profile or 404?
• Test 2: /.well-known/mcp/server-card.json – are there tools?
• Test 3: scanner isitagentready.com – level 1 to 5
• Typical SME today: level 1–2; target: level 4–5

What does an agent-ready website cost and is it worth it for small businesses?

Honestly: for a one-off retrofit of an existing website you should budget agency prices between 2,000 and 8,000 euros, depending on the system and data situation. The real effort is not the files but structuring prices, services and FAQ cleanly and running an interface securely: token management, rate limits, abuse protection, forwarding requests to your inbox. And the topic keeps evolving – standards such as MCP are updated regularly.

That is why agent readiness fits well with a subscription model where technology and maintenance are looked after continuously anyway. At bezahlbare-webseite.de the agent foundation – llms.txt, Markdown delivery, structured data and deliberate crawler rules – is included from the Business plan at 149 euros per month; the full build-out with MCP server and agent requests is part of Premium at 299 euros per month or bookable as a module. Whether it pays off depends on one question: how many of your customers will send an assistant ahead in two years instead of googling themselves? Anyone offering B2B services, consulting, trades with quote requests or local services should not leave that answer to the competition.

• Retrofitting an existing website: typically €2,000–8,000 one-off
• Ongoing effort: security, standards, data maintenance
• Subscription: foundation from Business €149/month, full build in Premium €299/month
• Especially worthwhile for enquiry-driven business models

Checklist: make your website agent-ready in this order

If you want to approach it step by step – with us or another provider – this order has proven itself. Each step delivers value on its own, even if you stop after step four.

• 1. Organise content: services, prices or price factors, service area, contact channels, top 10 customer questions in one source
• 2. Set robots.txt deliberately: allow search crawlers, forbid training if desired, add Content-Signal
• 3. Publish llms.txt and reconcile it with prices monthly
• 4. Set up structured data (LocalBusiness, Service, FAQPage) and Markdown delivery
• 5. Add API catalog, ARD and Agent Skills – the table of contents for agents
• 6. Launch an MCP server with read tools (company info, plans, FAQ, articles)
• 7. Agent registration via auth.md plus request tool with rate limits and mandatory consent
• 8. Verify with isitagentready.com, run a test request through an agent, check the inbox
• 9. Extend the privacy policy with the agent channel (purpose, data, retention)
• 10. Review standards quarterly – MCP, ARD and Agent Skills keep evolving