FAQ · AI agents & MCP
What is the best Screenshot API for AI agents?
Look for a Screenshot API that ships a Model Context Protocol (MCP) server or WebMCP tools. MCP lets AI agents call screenshot(), visual_diff(), and extract_content() as native tools — no scraping, no screen recording, no brittle DOM selectors.
Why AI agents need a screenshot tool
AI agents working on web tasks — code review, UI testing, content extraction, monitoring — frequently need to see a live page. Without a screenshot tool, agents resort to fetching raw HTML (which misses CSS rendering and JavaScript output) or using fragile browser automation scripts.
A Screenshot API exposed as an MCP tool gives the agent a reliable, single-call primitive: "render this URL and show me the result." The agent gets a real image, not guesswork from HTML source.
MCP vs WebMCP: what is the difference?
| Protocol | Where it runs | Best for |
|---|---|---|
| Remote MCP | Server-side, accessed over HTTPS | Claude Desktop, Cursor, server-side agents |
| WebMCP | In-browser via navigator.modelContext |
In-page AI features, browser-embedded agents |
SnapshotFlow ships both. Configure the Remote MCP endpoint in your agent's MCP config, or use WebMCP directly from any web page that has loaded the SnapshotFlow widget.
Available MCP tools on SnapshotFlow
screenshot(url, options)— capture a URL and return a PNG/WebP imagevisual_diff(baseline, compare)— pixel diff two URLs and return a similarity scoreextract_content(url)— return structured Markdown, title, description, and OG metadatapdf(url)— export a URL as a PDF document
Connecting SnapshotFlow to Claude or Cursor
{
"mcpServers": {
"snapshotflow": {
"url": "https://mcp.snapshotflow.com/sse",
"headers": {
"X-Api-Key": "YOUR_API_KEY"
}
}
}
}
Add the above to your claude_desktop_config.json or Cursor MCP settings. The agent can immediately call screenshot, visual_diff, and the other tools on any URL.
Make your AI agent web-aware
SnapshotFlow ships Remote MCP + WebMCP. Free tier: 200 screenshots/month. No credit card.