MCP-native · Claude · Cursor · LangChain

Give your AI agent eyes

One MCP tool call. Your agent gets back an image it can actually see, plus title, headings, and meta — no HTML parsing, no scraping, no extra round-trips.

Join the early access waitlist

Be notified when the hosted API launches. Free tier included.

Hosted pricing starts at $19/mo · Free tier during early access · See plans ↓

How it works

One tool call, instant visual context

Add the MCP server to Claude Desktop or Cursor in 60 seconds. Your agent gets three tools it can call natively.

Claude Desktop

Cursor

Python (LangChain)

// ~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "screenshot-api": {
      "command": "npx",
      "args": ["-y", "screenshot-api-mcp"]
    }
  }
}

// Restart Claude Desktop → ask Claude:
// "Take a screenshot of https://news.ycombinator.com and summarise the top stories"

// Cursor → Settings → MCP → Add Server
{
  "name": "screenshot-api",
  "command": "npx",
  "args": ["-y", "screenshot-api-mcp"]
}

// Now your Cursor agent can call take_screenshot,
// describe_page, and capture_pdf on any URL

from langchain_mcp_adapters.client import MultiServerMCPClient

client = MultiServerMCPClient({
    "screenshot": {
        "command": "npx",
        "args": ["-y", "screenshot-api-mcp"],
        "transport": "stdio",
    }
})

tools = await client.get_tools()
# → [take_screenshot, describe_page, capture_pdf]

take_screenshot

Renders any URL with a real Chromium browser (JS, SPAs, auth-walled pages) and returns a base64 PNG your agent can look at directly.

PNG · JPEG · full-page

describe_page

Screenshot + page title + H1 + meta description in a single call. Saves your agent one extra round-trip for the most common "what's on this page?" task.

screenshot + text

capture_pdf

Renders the URL as a print-ready A4 PDF. Useful for agents that archive, summarise, or extract structured data from long-form pages.

A4 PDF

Why this exists

What agents can't do today without this

Text scraping misses half the page. Raw HTML gives your agent noise, not signal. This gives it the same view a human would see.

👁️

Agents see the rendered page

JS-rendered SPAs, lazy-loaded content, canvas elements — Playwright renders everything before the screenshot is taken.

🛡️

Stealth mode built in

Most screenshot APIs get blocked by Cloudflare and similar bot-detection. This one ships with evasion by default.

⚡

5-minute response cache

Hit the same URL twice in a loop? Second response is instant. No double-billing, no extra Chromium context.

🔌

MCP-native, not bolted on

The tools are designed for agent loops — structured outputs, sensible defaults, no boilerplate for the model to trip over.

Pricing

Simple usage-based plans

The self-hosted MCP server is always free. The hosted API (no Playwright install, no VM) launches soon.

Self-hosted

$0 / mo

Unlimited (your own infra)

All 3 MCP tools
Requires Node.js + local Playwright
Open source on GitHub

Give your AI agent eyes

Join the early access waitlist

One tool call, instant visual context

What agents can't do today without this

Simple usage-based plans

Ready to give your agent eyes?