Connect Snapshot Site to your MCP client
This page is the setup path: which endpoint to point your client at, how each connection mode authenticates, and what the three tools return. For what the MCP server does and why, read the MCP server overview.
https://mcp.snapshot-site.com/mcpSNAPSHOT_SITE_API_KEY=ss_live_xxxThe hosted endpoint is live and published to the Model Context Protocol registry as com.snapshot-site/screenshots. A local stdio mode is available if you would rather run the server yourself with an API key.
Tools exposed to the client
All three are annotated read-only, idempotent, and open-world, so an agent can retry them safely and a client can reason about their side effects before calling.
| Tool | Name | Endpoint | What it does |
|---|---|---|---|
| Tool | screenshot | POST /api/v2/screenshot | Capture a page as PNG, JPEG, WebP, PDF, base64, or raw HTML, with viewport, full-page, and cookie-banner controls. |
| Tool | analyze | POST /api/v3/analyze | Capture a page and return an AI summary, topics, a heading outline, SEO metadata, and quality checks the agent can branch on. |
| Tool | compare | POST /api/v3/compare | Diff two pages, or a page against a baseline image, and return a pixel mismatch score plus a diff image. |
Hosted connection, no local process
- 1
Add a custom MCP connector in your client and point it at
https://mcp.snapshot-site.com/mcp. - 2
Sign in when the client opens the OAuth prompt. The server resolves your account from the session — there is no API key to paste.
- 3
Approve access. The three tools then appear in the client's tool list as
screenshot,analyze, andcompare. - 4
Ask the assistant for what you want — a capture of a URL, an analysis of a page, a comparison of two versions — and it selects the tool.
- 5
Prefer to run it yourself? Launch
npx -y @snapshot-site/mcpover stdio withSNAPSHOT_SITE_API_KEYset instead of using the hosted endpoint.
Limits worth knowing before you build
Usage is metered like any API call
A tool call spends the same quota as a direct request to the endpoint. Nothing is billed differently because it arrived through MCP.
Client support varies
The hosted endpoint speaks the streamable HTTP transport and the local package speaks stdio. A client that supports neither cannot connect — check which transports yours implements before wiring it.
Local mode requires Node.js 20.9 or later
Only the stdio path is affected. The hosted endpoint has no local runtime requirement at all.
The tools return data, they do not store it
There is no server-side history of captures to query. If an agent needs to compare against last week, the baseline has to live somewhere your workflow controls.
Tool selection is the client's decision
The tools carry titles, category metadata, and example intents to help, but which one gets called for a given prompt is up to the client's model, not the server.
MCP integration FAQ
Which MCP clients can connect to Snapshot Site?
Any client that speaks the streamable HTTP transport can use the hosted endpoint, and any client that speaks stdio can run the local package. The server has been tested with Claude Desktop and Cursor over stdio, and against the hosted endpoint over streamable HTTP.
Do I need an API key for the hosted MCP server?
No. The hosted endpoint uses OAuth — you sign in and the server resolves your account from the session, so no key ends up in your client config. An API key is only needed for local stdio mode.
Does going through MCP cost more than calling the API?
No. A tool call counts against the same plan quota as a direct API call. There is no separate MCP metering.
What is the registry entry for?
The server is published to the Model Context Protocol registry as com.snapshot-site/screenshots, so clients that browse the registry can discover it instead of needing the endpoint URL pasted in by hand.
Can an agent monitor a page for changes over time?
It can compare any two things you give it, but the server keeps no history. Persistent monitoring needs a baseline stored on your side — or a scheduled workflow, which is what the n8n and Zapier integrations are for.
Related resources
Give your agent a browser
Point your MCP client at the hosted endpoint, sign in, and your assistant can capture, analyze, and compare pages on its own. The free plan includes 50 captures per month.