Screenshot Scout

Screenshot Scout MCP Server

Use Screenshot Scout from an MCP client to capture HTTP or HTTPS webpages as images or PDFs.

This server exposes one tool, capture_screenshot. It supports full-page and element captures, device and viewport controls, location selection, page interaction and blocking options, image sizing and quality, PDF layout, caching, temporary result URLs and eligible MCP image content.

What you need

• A Screenshot Scout account and an access key from the API keys page.
• Node.js 22 or newer for npm/stdio installation. Claude Desktop's MCPB runtime is bundled by Claude.
• The optional secret key only when your selected API key requires signed Screenshot Scout requests.

Every capture uses your Screenshot Scout account and is subject to its plan, quota, and rate limits.

Local stdio with npm

Start with this local stdio configuration:

{

"mcpServers": { "screenshotscout": { "command": "npx", "args": ["-y", "@screenshotscout/mcp"], "env": { "SCREENSHOTSCOUT_ACCESS_KEY": "YOUR_ACCESS_KEY" } } } }

If the access key requires request signing, add the secret locally:

"SCREENSHOTSCOUT_SECRET_KEY": "YOUR_SECRET_KEY"

Keep personal configuration files out of source control. Credentials are process environment values, not tool arguments. See client-specific copy-paste configurations for Claude Desktop, Claude Code, Cursor, VS Code/GitHub Copilot, Devin, and Cline.

Run from a source checkout

npm ci

npm run build

Point the client at the absolute path to dist/stdio.js with node, and provide the same environment variables shown above.

Claude Desktop MCPB

To install the Claude Desktop extension:

1. Download screenshotscout-mcp- .mcpb from that version's GitHub release.
2. In Claude Desktop, open Settings → Extensions → Advanced settings and choose Install Extension….
3. Select the downloaded file.
4. Enter the required access key. Enter the secret key only for an API key that requires signed requests.

Claude Desktop treats both fields as sensitive settings. The v0.3.0 MCPB supports Windows.

Hosted Streamable HTTP

The hosted API-key endpoint is live at:

https://mcp.screenshotscout.com/mcp/api-key

It is intended only for clients that can attach a static HTTP header:

Authorization: Bearer YOUR_ACCESS_KEY

The endpoint accepts an access key only. Never send a Screenshot Scout secret key to it, and never put either key in the URL or a tool argument. Clients that cannot attach a static Bearer header cannot use this endpoint.

API keys that require request signatures must instead use local stdio or MCPB, or use a dedicated unsigned access key for the hosted endpoint.

Local stdio with Docker

Build the production image from a source checkout:

docker build --tag screenshotscout-mcp:local .

Pass credentials from the local environment and keep stdin attached for MCP stdio traffic:

docker run --rm -i --init --cap-drop=ALL --security-opt=no-new-privileges --read-only \

-e SCREENSHOTSCOUT_ACCESS_KEY \ -e SCREENSHOTSCOUT_SECRET_KEY \ screenshotscout-mcp:local

SCREENSHOTSCOUT_SECRET_KEY remains optional. The image runs as an unprivileged user and contains only the compiled stdio server and its production dependencies. It declares no port or container health check: an MCP client owns the stdio process and verifies readiness by completing MCP initialization. The image and its Docker MCP Catalog metadata in docker-mcp-catalog.yaml are local preparation; no public image is implied by these commands.

Tool: capture_screenshot

capture_screenshot sends one capture request for the supplied URL and options. The target webpage is external, and its returned content must be treated as untrusted.

Inputs

Only url is required. Captures use a 1280×720 viewport by default. When no format is specified, the tool returns JPEG at quality 60. resultMode defaults to "auto".

When both output dimensions are provided, their product cannot exceed 64,000,000 pixels. PDF margins accept non-negative values in px, in, mm, or cm. Inputs for another format are ignored: imageQuality applies only to JPEG and WebP, and PDF inputs only to format: "pdf". A PDF always contains the whole page, so fullPage, selector, imageWidth, and imageHeight are ignored for PDF.

Results

• PNG, JPEG, WebP, and GIF may be included as MCP image content when resultMode is auto, the MIME type is eligible, dimensions are known and at most 8,000 pixels per side, raw data is at most 5 MiB, and the complete serialized result fits the current 128,000-byte server limit.
• A capture that is not eligible to embed remains successful and returns its temporary URL plus an actionable omission reason.
• TIFF is URL-only.
• PDF bytes are never embedded. A PDF result includes safe text and structured metadata, plus a resource link when Screenshot Scout provides a result URL.
• resultMode: "url_only" omits image bytes for every format.

MCP clients control whether returned image content or resource links are displayed or made available to a model.

Structured metadata can include screenshotUrl, screenshotUrlExpiresAt, cacheStatus, format, mimeType, imageWidth, imageHeight, inlineImageIncluded, and inlineImageOmissionReason.

Treat result URLs as sensitive, temporary links and respect their reported expiry.

Example prompts

• “Capture https://example.com as a full-page PNG in dark mode. Return only a URL.”
• “Take a 1280×720 JPEG screenshot of https://example.com/pricing, block cookie banners and ads, and use quality 80.”
• “Create an A4 PDF of https://example.com/report with backgrounds enabled and 10 mm margins.”

Privacy and security

The server sends the target URL and selected capture options to Screenshot Scout, which loads the target website. Review the Screenshot Scout privacy policy before capturing private or regulated material.

• Do not capture pages you are not authorized to access.
• Do not paste credentials into prompts, tool inputs, URLs, issue reports, or logs.
• Keep local access and secret keys in client-managed secret storage or private environment configuration.
• The local stdio server adds no telemetry. Application logging for the hosted service is limited to request method, response status, duration, and sanitized unexpected errors. It is designed not to include credentials, target URLs, screenshot URLs, request or response content, or image bytes.
• Review each target and capture request before allowing tool use. The tool is open-world, consumes quota, and interacts with an external website.
• Report vulnerabilities privately as described in SECURITY.md.

Development

npm ci

npm run format:check npm run lint npm run typecheck npm test npm run metadata:check npm run registry:validate npm run mcpb:validate npm run mcpb:pack

License

MIT © Oleksii Velykyi