For developers building AI agents

Your agent can't see the web.
Give it eyes.

AgentEyes is a screenshot + page-data API that's MCP-native from day one. One call from Claude or Cursor returns a screenshot, the accessibility tree, and extracted text — no Playwright cluster to babysit, no hand-written wrappers.

Get early access See how it works

Thirty seconds to working

Install the free MCP server, point Claude Desktop or Cursor at it, done.

# install the open-source MCP server
npx -y agenteyes-mcp

# then in Claude, just ask:
"Screenshot stripe.com/pricing and tell me what the middle tier costs"

The agent gets back a PNG, the accessibility tree, and the page text — in one call.

MCP-native, not bolted on

One-line install into Claude Desktop or Cursor. Your agent calls screenshot_url like any other tool — because it is one.

One call, three things back

Screenshot + accessibility tree + extracted text. Agents don't need pixels alone; they need to understand the page.

Free to self-host

The MCP server is MIT open source — run it yourself forever. The hosted API is usage-based with a generous free tier for when you don't want the ops.

Yeah, ScreenshotOne ($17/mo) already exists — and PageBolt has an MCP server too. We're not in the cheaper-pixels business. The bet is that agent builders want page understanding (screenshot + a11y tree + text in a single call), not just PNGs. If all you need is a screenshot on a cron, the incumbents are honestly fine.
Free self-hosted, forever

The MCP server is MIT licensed. Run it on your own machine, no account needed.

Hosted API — usage-based, generous free tier

For when you'd rather not babysit browsers. We're validating demand before building it —
join the list and tell us which camp you're in. Nothing is charged today.

Get early access

We're validating the hosted version before building it. Join the list.

You're on the list.

Thanks — we'll email you when the MCP server and early API access are ready.

No spam. One email when it ships, that's it.