Find buttons and links
Roles, names, and states come back as JSON. Agents skip decorative divs and focus on actionable nodes.
Accessibility tree for browser agents
Let agents navigate like users do
Computer-use agents need structure, not HTML soup. The /a11y endpoint returns the accessibility tree after a real render so agents click the right elements.
Works with your agent stack
Use cases
Structure agents can act on
DOM snapshots are huge and brittle. Accessibility trees expose interactive elements with stable roles and labels your agent can target.
Post-render accuracy
The tree reflects the page after JavaScript runs, not the empty shell from a static fetch.
Pair with screenshots
Combine /a11y with /screenshot when the agent needs both visual context and a navigable element map.
Technical approach
Accessibility tree, not DOM dump
We render the page in a headless browser, then serialize the ARIA tree. Agents get a compact JSON graph tuned for action selection.
Full ARIA coverage
Roles, labels, descriptions, states, and hierarchy for every accessible node on the page.
JS-aware
SPAs and client-rendered apps return the same tree a screen reader would see after hydration.
Agent-sized payload
The tree is smaller than raw HTML and aligned to how assistive tech interprets the page.
Integration
Wire into browser agent loops
Fetch tree
POST the URL to /v1/a11y. Parse the JSON tree in your agent loop.
View docs →Select targets
Filter nodes by role (button, link, textbox) and accessible name to pick the next action.
Agent tools →Verify with screenshot
Capture a screenshot after each action to confirm the agent clicked the intended element.
Screenshot API →001How is this different from /content?
/content returns HTML. /a11y returns the semantic accessibility tree as JSON, optimized for agent action selection.
002Does it work on SPAs?
Yes. We render with a headless browser first, so client-side routes and dynamic widgets appear in the tree.
003Can I use this with MCP?
Yes. The a11y tool is available in the CaptureWeb MCP server alongside screenshot and markdown.
Explore more
Web capture for every workflow
AI & agents
RAG pipelines
Feed your vector database with fresh, clean web content stripped of boilerplate.
Learn more →AI & agentsAI agents & MCP
Give your AI agent live eyes on the web. One MCP install, ten tools.
Learn more →Data extractionBrand extraction
Turn any company domain into a structured brand profile.
Learn more →Give agents a navigable map
Fetch the accessibility tree from any public URL in one call.