Give your AI agent a screenshot tool. aiSnapApi exposes capture_webpage over the Model Context Protocol — works in Claude, Cursor, Windsurf, Claude Code, and any MCP-compatible client. One stateless HTTPS endpoint, no install.
No npm package, no Docker, no stdio bridge. Add aiSnapApi to your client's MCP config:
# Cursor / Windsurf / generic MCP clients (~/.cursor/mcp.json or equivalent):
{
"mcpServers": {
"aiSnapApi": {
"url": "https://api.aisnapapi.com/v1/mcp",
"headers": { "Authorization": "Bearer YOUR_API_KEY" }
}
}
}
# Claude Code one-liner:
claude mcp add --transport http aiSnapApi https://api.aisnapapi.com/v1/mcp \
--header "Authorization: Bearer YOUR_API_KEY"
# Claude.ai / ChatGPT — easiest: paste this as the custom connector URL
# (use this if the connector UI doesn't expose a headers field):
https://api.aisnapapi.com/v1/mcp?api_key=YOUR_API_KEY
# Or, if the UI accepts custom headers:
# URL: https://api.aisnapapi.com/v1/mcp
# Auth: Bearer YOUR_API_KEY
Replace YOUR_API_KEY with the key you got from the customer portal (free tier needs no card).
One tool, capture_webpage, mirrors every parameter of the REST endpoint:
| Parameter | Type | Default | Description |
|---|---|---|---|
url | string | — | Required. http or https. |
format | enum | png | png, jpeg, webp, pdf |
width | integer | 1280 | 1–3840 |
height | integer | 720 | 1–2160 (ignored when full_page=true) |
full_page | boolean | false | Capture the entire scrollable page |
delay | integer | 0 | Milliseconds to wait after load (0–10000) |
dark_mode | boolean | false | Emulate prefers-color-scheme: dark |
css | string | — | Custom CSS to inject before capture |
quality | integer | 80 | JPEG / WebP quality 1–100 (ignored for PNG) |
The tool returns an inline image content block (PNG / JPEG / WebP) or a resource block (PDF), plus a text block with the current month's usage so the agent can self-throttle.
Agent calls: capture_webpage(url="https://news.ycombinator.com", full_page=true)
Returns: a PNG of the page plus a usage summary ("47/1000 this month, 953 remaining"). The agent reads the image, identifies the top story, and answers.
Agent calls: capture_webpage(url="https://aisnapapi.com/#pricing", format="pdf")
Returns: a PDF resource block the agent can save to disk or attach to a message.
Most MCP screenshot servers ship as npm packages your client launches as a subprocess. That works, but it means every developer has to install Node, install the package, and keep it updated. The HTTP transport skips all of that — your agent just opens a connection to a URL, sends JSON-RPC, and gets results.
The endpoint is stateless: each tool call is a self-contained POST, no SSE streaming, no session tokens. It rides the same load-balanced infrastructure as the REST API, so you get the same latency and the same uptime.
Start with 100 captures per month, no card required. Upgrade from the portal when you need more — change or cancel anytime.