mcp.screenshotink.com

Screenshot MCP server for AI agents

Claude · Cursor · ChatGPT · Windsurf

Five tools your agent can call to capture, batch, compare and audit any page — and see the result in the same turn. No local Chrome, no Puppeteer maintenance, works in CI and cloud sandboxes.

Get free API key API docs

100 captures/month free · the same key works over MCP and REST

✓ MCP server live Claude DesktopClaude CodeCursorWindsurf GitHub ↗

claude_desktop_config.json

{
  "mcpServers": {
    "screenshotink": {
      "command": "npx",
      "args": ["@screenshotink/mcp"],
      "env": { "SCREENSHOTINK_API_KEY": "sk_live_…" }
    }
  }
}

terminal

# Claude Code — one command:
claude mcp add screenshotink -- npx -y @screenshotink/mcp --key sk_live_…

# or run anywhere: npx -y @screenshotink/mcp --key sk_live_…
# stdio transport · zero install · Node 18+

streamable HTTP

https://mcp.screenshotink.com/mcp

# auth: Authorization: Bearer sk_live_…
# for cloud agents & hosted runtimes — nothing to run

Custom connector

Add ScreenshotInk to your AI tool

We're a standard remote MCP server — so you can connect it today, as a custom connector, without waiting for any directory or store listing. Point your tool at one URL and sign in, or run it locally with your API key.

Remote · recommended

https://mcp.screenshotink.com/mcp

Streamable HTTP. Sign in with your ScreenshotInk account on first use (OAuth) — nothing to install, works in cloud agents.

Local · stdio

npx -y @screenshotink/mcp --key sk_live_…

Runs on your machine with your API key. Node 18+. Best for desktop apps that take a command.

Claude.ai (web) and Claude Desktop both take a custom connector by URL — with sign-in handled over OAuth.

1Open Settings → Connectors and click Add custom connector.
2Name it ScreenshotInk and paste the remote URL:

https://mcp.screenshotink.com/mcp

3Click Add, then Connect and sign in with your ScreenshotInk account. The five tools appear in the 🔌 menu.

Custom connectors are available on Claude Pro, Max, Team and Enterprise. On Claude Desktop you can instead drop the local claude_desktop_config.json block from the top of this page; in Claude Code run claude mcp add --transport http screenshotink https://mcp.screenshotink.com/mcp.

Cursor connects to remote MCP servers by URL and handles the OAuth sign-in for you.

1Open Settings → Tools & Integrations → New MCP Server (or edit ~/.cursor/mcp.json).
2Add the server:

~/.cursor/mcp.json

{
  "mcpServers": {
    "screenshotink": {
      "url": "https://mcp.screenshotink.com/mcp"
    }
  }
}

3Save, reload, and approve the sign-in. Prefer a key? Swap "url" for a "headers": { "Authorization": "Bearer sk_live_…" } entry alongside it.

GitHub Copilot's agent mode reads MCP servers from a workspace file.

1Create .vscode/mcp.json (or run MCP: Add Server → HTTP from the command palette).
2Add the server:

.vscode/mcp.json

{
  "servers": {
    "screenshotink": {
      "type": "http",
      "url": "https://mcp.screenshotink.com/mcp"
    }
  }
}

3Press Start on the server in the MCP view and sign in when prompted. Open Copilot Chat in Agent mode to use the tools.

Windsurf's Cascade reads MCP servers from its config file.

1Open Cascade → MCP → Configure (or edit ~/.codeium/windsurf/mcp_config.json).
2Add the server:

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "screenshotink": {
      "serverUrl": "https://mcp.screenshotink.com/mcp"
    }
  }
}

3Hit Refresh in the MCP panel and authenticate. Tools show up under the 🔨 hammer.

Tool reference

Five tools, one key

Same parameters and defaults as the REST API. Every result returns the image inline, so the agent can reason about pixels immediately.

take_screenshot

{ url, width?, full_size?,
  format?, no_ads?, lazy_load? }

When an agent needs to verify a deploy, check a layout, or read a page visually.

bulk_screenshots

{ urls: string[≤20],
  width?, full_size?, format? }

When an agent audits a whole funnel or a competitor set in one call.

compare_screenshots

{ url_a, url_b | image_a, image_b,
  threshold? }

When an agent checks staging vs production, or before vs after a change.

run_lighthouse

{ url, categories?:
  ["performance","a11y","seo"] }

When an agent needs scores plus the rendered page to explain why.

capture_sitemap

{ sitemap_url, limit?: ≤50,
  width?, format? }

When an agent inventories a site visually — every page the sitemap lists.

Captures over MCP count against the same quota as the REST API. One key, one meter.

See quotas →

In practice

A real agent session

MCP session claude · mcp.screenshotink.com

Compare our pricing page on staging vs production — did the new plan cards ship correctly?

▸ compare_screenshots { "url_a": "https://acme.com/pricing", "url_b": "https://staging.acme.dev/pricing" }

diff · A ↔ B · threshold 0.1

Diff by ScreenshotInk · 4.7% changed

Changed regions highlighted — labels included for non-color reading

The plan grid changed from 3 to 4 cards on staging — the new "Scale" card is present and highlighted. Card widths shrank from 84px to 68px equivalents; spacing is consistent. 4.7% of pixels differ, all within the plan grid region. The deploy looks correct and intentional. ✓

Why hosted

Works where local browsers don't

CI pipelines

No headless Chrome install, no flaky browser binaries in your runners. One HTTP call from any job.

Cloud agents

Hosted runtimes can't launch browsers. The remote MCP endpoint gives them eyes without a single dependency.

Zero maintenance

We patch Chromium, rotate IPs, handle lazy-load and cookie walls. Your agent just asks for pixels.

Give your agent eyes in 30 seconds

Free key, 100 captures/month. One npx command or the remote URL — live now. No card required.

Get free API key Setup docs