Level 1 to Level 4
isitagentready.com Level
From "Basic Web Presence" to "Agent-Integrated"
1,391
Pages With Markdown Views
Every public page answers with a .md view and Accept: text/markdown
Passed
Cache-Separation Gate
Markdown and HTML stayed separate in the production cache, three rounds out of three
6
MCP Tools Shipped
Five read-only tools plus discovery-call booking; first real production booking still to verify
10
Top Pages Rewritten Answer-First
Answer-first rewrites of our highest-priority pages
Where Things Stood Before
We tell HubSpot teams that more of their buyers ask an AI assistant before they ever visit a website. Fair enough. So how did our own site look to one of those assistants?
Not great. On October 1, 2026, we ran sidekickstrategies.com through isitagentready.com, Cloudflare's agent-readiness checker. It came back Level 1, "Basic Web Presence," with 3 checks passing, 13 failing and 6 neutral.
The gaps were specific:
- No Link headers on any page, so an agent had no pointer to a cleaner version of the content.
- No markdown negotiation. Asking the homepage for text/markdown returned HTML anyway, and a .md version of a service page returned a 404.
- No agent skills index and no MCP server for an agent to call.
- Our own robots.txt disallowed /api/. The MCP server we planned would live at /api/mcp, so the site would have blocked it on day one.
There was a second challenge, and it was honesty. The research we did before writing any code turned up a stat worth sitting with: Profound's analysis of 27 million AI citations found only 4% point at a brand's own website. Third-party presence on review sites, YouTube and Reddit matters more. So we went in with clear eyes. This work is cheap, correct, and makes a site legible. It isn't the growth lever on its own.
“More of your buyers ask an AI assistant before they ever visit a website, and we want our answers to be the clear ones when that happens.”
What We Built Together
We split the build into phases and shipped each one to production.
A markdown version of every public page. All 1,391 public pages now answer with a .md view and respond to Accept: text/markdown. Every HTML page advertises its markdown alternate in a Link header. Publishing in Sanity purges the cached copies, and an agent:check drift guard rides along.
Discovery. robots.txt now allows the MCP endpoint, carries Content Signals, and opts out of AI training. We generate llms.txt and published our H.E.A.R.T. skill so a developer can install it with npx skills add.
A legible offer. An agent can't quote what it can't read, so prices now live in readable text. We rewrote our ten top pages answer-first.
An MCP server. /api/mcp ships five read-only tools for things an agent can't get by fetching HTML, like cohort dates and real discovery-call availability. A sixth tool books a discovery call, with rate limits, a disposable-email block, and a HubSpot workflow that sets an "AI Agent" lead source on agent bookings that arrive without one.
A page for humans. The new /agents page explains all of it in plain English and lists the tools straight from the server.
Just as important is what we chose not to build, and why:
- No Cloudflare proxy. Vercel's docs say its Bot Protection doesn't work behind a reverse proxy, and Vercel already verifies Web Bot Auth natively. Cloudflare stays DNS-only.
- No Stripe Payment Links. Our checkout runs on HubSpot Payments.
- No buy button for retainers. A strategy call stays the next step.
- No seat counts. Not on pages, not in tools.
- No agent-run checkout or recurring billing (ACP, UCP checkout, x402). In 2026, an agent signing a human up for a monthly retainer, unattended, end to end, isn't supported by the payment specs. The agent gets the price, dates and links. The human completes the purchase.
- No FAQPage schema. Google removed FAQ rich results in May 2026. We kept the FAQ content as visible text.
- llms.txt, with no expectations. It's cheap to publish, but Ahrefs found 97% of llms.txt files they sampled got zero requests.
The Transformation
The rescan after the /agents deploy moved sidekickstrategies.com from Level 1, "Basic Web Presence," to Level 4, "Agent-Integrated." Link headers, markdown negotiation, Content Signals and agent skills all flipped to passing.
The markdown layer cleared its hardest test on production. Serving markdown and HTML on the same URL risks the cache handing one to the other. Across three rounds of paired requests, markdown came back as markdown and HTML as HTML every time, because Vercel keys its cache on the Accept header.
The MCP server was verified on production and returned real HubSpot booking slots. The booking tool is live, and we're still waiting on its first real production booking to confirm it end to end.
A few checks still fail, by choice: DNS-AID, API catalog, OAuth discovery, auth.md, a /.well-known MCP card (that spec is still an unmerged draft), an A2A card, WebMCP and ARD. Each one is a decision we made, not a gap we missed.
What we can't tell you yet is whether agents are showing up. Traffic and agent-usage numbers are pending a two-week measurement window that ends October 16, 2026. After that we'll report .md fetches, negotiated fetches, MCP calls by tool, assistant referrers, and a Search Console check that no .md URLs got indexed. We won't guess at those numbers before we have them.
Want the same treatment for your site? That's something we teach. See how we make a site agent-ready.
Worked with us? We'd love to tell your story.


