Built for
AI agents.
Your agent needs to see the web. Shotbot delivers reliable screenshots via a clean JSON API, a native MCP server, batch, and async callbacks. When it needs to work for real.
Use cases for AI agents
Web research agents
Your agent can read the current visual state of any page without running a headless browser itself. Pass the screenshot URL or raw file to a vision model for analysis.
Visual regression in CI
On each deploy, batch-capture your key pages. Compare pixel-by-pixel or feed to a multimodal model. Structured errors (error codes, HTTP status) make it easy to script.
Document archiving
Capture as PDF with a WebP preview. Keep it private (the default over the CLI and MCP) for a download link that is never indexed on the CDN and auto-expires, ideal for confidential documents.
Price & content monitoring
Submit 500 product or competitor pages in one batch request.
Receive each result via callback_url as it completes.
Your agent is notified, it does not poll.
Report generation
Batch-capture dashboard URLs as PDF, embed in automated reports. Full-page captures up to 30 000 px, with dark mode or custom CSS selector clipping.
Data pipelines
OpenAPI 3.1 spec at api.shotbot.net/openapi.yaml. Import directly into any agent framework, LangChain tool, or OpenAPI client generator.
MCP server | native agent integration
The Shotbot MCP server exposes four tools directly inside
Claude Code, Claude Desktop, Cursor, Windsurf, and any MCP-compatible runtime.
Your agent calls capture like any other tool, no wrapper code needed.
> Analyse this competitor's pricing page and compare it with ours ● I'll capture the pricing page now. capture(url: "https://www.permalink.fr/tarifs/", viewport_width: 1280) ✓ rendered in 2s · https://static.shotbot.net/3/3f/3f9eC4d5…jpg ● The page shows three plans: Starter €29, Pro €79, Enterprise on request. Here's how they compare with your current pricing... > _
Installation
On Claude Code, one command registers it. No Node.js or npm needed.
claude mcp add --scope user shotbot --transport http "https://api.shotbot.net/mcp?key=your-key"
For Claude Desktop, Cursor, Windsurf and other MCP-compatible runtimes, add to your config file:
{
"mcpServers": {
"shotbot": {
"url": "https://api.shotbot.net/mcp?key=your-key"
}
}
}
Why Shotbot for AI workflows
- API-first: JSON in, JSON out. Structured error codes (
pro_required,concurrency_limit,invalid_url) are safe toswitchon. - Async by design:
callback_urldelivers each result via webhook | your agent is notified, it does not poll in a loop. - Private by default: over the CLI and MCP, captures skip the CDN entirely. The file is streamed on demand (fetch it repeatedly) and auto-expires; pass
cdn=truefor a permanent public URL. - OpenAPI 3.1: import openapi.yaml into LangChain, LlamaIndex, or any OpenAPI-aware agent framework.
- Machine-readable docs: llms.txt at the domain root; the MCP tool descriptions are LLM-optimized.
- Hosted in France: GDPR-compliant, no third-party cloud, 18 years of uptime.
Get started
Free account includes 200 screenshots/month. No credit card required. API key available immediately after sign-up.
Related use cases
Visual monitoring
Scheduled captures (6h, 24h, 7d, monthly) to feed your change-detection pipelines. Each run generates a distinct timestamped snapshot.
Learn more →Legal archiving
Each screenshot is a timestamped and immutable snapshot. Ideal as visual evidence or to track the evolution of a site over time.
Learn more →Export to S3
Push each screenshot directly into your bucket via callback. Works with AWS, Scaleway, OVH, Cloudflare R2, Backblaze B2.
Learn more →