screenroveSCREENS WORTH STUDYING
← Field notes
Pattern study

AI-ready documentation: docs pages now written for coding agents

AI-ready documentation from real product screens: Stripe's Copy for LLM and View as Markdown, Raycast's Ask or search, Clerk's SKILL.md and Eraser's MCP setup.

Every section of Stripe's API reference now carries three actions that no human reader strictly needs: "Ask about this section", "Copy for LLM" and "View as Markdown". That small row is the clearest sign in our library of AI-ready documentation: docs pages that assume the next reader may be a coding agent, or a person about to paste the page into one.

The sample here is small. The blog pages wall has 18 screens from 12 products, and only 10 of them are docs or help pages (the 9 tagged help / docs plus Raycast's docs-hosted changelog). Those 10 come from 6 products. Two of the six, Stripe and Raycast, put an AI or copy affordance on every docs screen we captured: 6 of the 10 screens. So we widened the lens to the rest of the library, where the same idea shows up on landing pages and onboarding steps.

Stripe: the section is the unit

On the API reference, the action row repeats on every section header we saw: Introduction, Authentication, Customers, The Customer object. It is set in small grey type, so it never competes with the heading. Code panels add their own pair of icons, copy and a sparkle, so a single snippet can be lifted or questioned on its own.

Stripe: API reference sections each headed by Ask about this section, Copy for LLM and View as Markdown
Two sections, two identical action rows. The code panels carry their own copy and sparkle icons, and the human feedback line ("Was this section helpful?") works at the same section level. Captured 24 September 2026 · DesktopExplore reference ↗

Two details matter more than the buttons. The unit is the section, not the page. An agent working on customers wants the Customer object and its endpoints, not the whole reference, and "Copy for LLM" hands over exactly that slice. And the human feedback prompt ("Was this section helpful? Yes No") sits at the same granularity, so the docs team can see which sections people needed help with.

The search box is paired too. In the API sidebar, "Find anything" sits beside "Ask AI"; on the docs home, "Search" sits beside "Ask AI" in the header. All four Stripe docs screens on the wall, including the Customers endpoint list and the APIs & SDKs menu, show that pair.

Raycast: one Copy button and a search that answers

Raycast's developer docs run on GitBook ("Powered by GitBook" sits at the foot of the sidebar). The header search placeholder reads "Ask or search…", merging the two boxes Stripe keeps apart. Both captured pages have a Copy split button beside the title: the Introduction and on the changelog alike.

Raycast: GitBook docs Introduction page with a Copy split button beside the title
One split button, Copy plus a dropdown, at title level. Out of frame, the header search reads "Ask or search…" with a ⌘K hint. Captured 24 September 2026 · DesktopExplore reference ↗

It is a lighter version of Stripe's row: one control per page instead of three per section. Page-level copying is fine for guides; for long references, section-level copying hands an agent far less noise.

The same changelog shows where this is heading: Raycast's 2.5.0 notes tell extension authors how to declare an MCP server and bundle Agent Skills, each in a SKILL.md file. The docs are teaching developers to write docs for agents.

Beyond the docs site: one line an agent can read

Elsewhere in the library, products compress their docs into one fetchable line and put it where a button would go.

Clerk: hero with a Build with agents chip holding the line Add Clerk auth to my app: clerk.com/SKILL.md
The hero has no button. Its only action is a copyable instruction that points an agent at one Markdown file. Captured 2 October 2026 · DesktopExplore reference ↗

Clerk's feature page is the boldest case: under the headline, a "Build with agents" chip holds Add Clerk auth to my app: clerk.com/SKILL.md and a copy icon. There is no other hero action. Netlify's home does the same with "Add Netlify to your coding agent" above a copyable fetch https://netlify.ai. And the footer of Product Hunt's launch guide links llms.txt between Advertise and Contact.

Rive: hero box reading Install the Rive CLI from rive.app/cli. Let's build a 3D city. with a COPY button
"Start with your coding agent. No account needed to try it." The quickstart is a two-sentence prompt with the CLI URL inside it. Captured 22 October 2026 · DesktopExplore reference ↗

Rive goes one step further and writes the prompt for you: "Install the Rive CLI from rive.app/cli. Let's build a 3D city." with a COPY button, then "Then open it in the Rive Editor to make it yours." The docs URL lives inside the prompt, so the agent finds the instructions on its own.

Setup steps written as prompts and commands

Onboarding is where this pattern turns into a default. Amplitude's "Let's get data in!" leads with an "AI prompt" card tagged "Recommended": "Run this custom prompt in an AI coding tool like Cursor or Claude to set up Amplitude", with "Open in Cursor" and "Copy Prompt".

Amplitude: setup card tagged Recommended with a prompt, Open in Cursor and Copy Prompt buttons
The prompt is tagged "Recommended" and toggles with a plain Snippet tab; integrations and "Invite Team" sit below "or set up another way". Captured 21 October 2026 · DesktopExplore reference ↗

Eraser makes MCP the docs: pick one of 10 clients (Claude Code, Cursor, VS Code, GitHub Copilot and others) and the card shows the exact command, such as claude mcp add --transport http eraser https://app.eraser.io/api/mcp, with a Copy chip. The full setup guide is still there, as a text link.

Eraser: Connect your MCP client step with 10 client tiles and a claude mcp add command with a Copy chip
Ten client tiles, one exact command per client, and the long-form guide demoted to a text link. This is step 4 of 5 in onboarding. Captured 22 October 2026 · DesktopExplore reference ↗

More are on the integration connect view and in an MCP search.

What to add to your docs

How we counted

All screens were captured in desktop browsers with research accounts in September and October 2026. A vision model recorded each screen's attributes, and every claim here can be checked on the linked screen pages. With 10 docs screens from 6 products, this is a pattern study, not a census, and we capture what we saw, not usage or conversion data; we cannot tell you how often anyone presses "Copy for LLM".

Screenrove practises this too: we serve an llms.txt and an MCP server, so you can connect the Screenrove MCP server and ask your agent for docs pages with AI actions before you design your own.