feat(server): let agents save browser screenshots

[adinschmidt]

Agents can save a recording of a collaborative browser tab, but they have no way to save a still. preview_snapshot returns a downscaled PNG inline for the agent to look at, and the desktop camera button writes a file, but that capture was never exposed over MCP — so an agent that just wants to hand the user a still image has to record a video instead.

This adds a preview_screenshot tool that reuses the existing desktop capture and returns the saved artifact's path, mirroring preview_recording_stop:

preview_screenshot { tabId? } → { id, tabId, path, mimeType: "image/png", sizeBytes, createdAt }

Artifacts land in the same browser-artifacts directory as recordings and the camera button, so reveal and copy-path work unchanged.

The tool returns a path rather than image data on purpose: preview_snapshot already covers "let the agent see the page", and re-encoding a full-resolution PNG into the transcript on every capture would be a lot of tokens for an evidence artifact the human is meant to open.

"screenshot" joins PREVIEW_AUTOMATION_OPERATIONS but not the V1 list, so desktop hosts predating this change are filtered out by the broker and get a capability error rather than an operation they cannot serve.

One inherited constraint worth knowing: capture needs a composited tab. This calls the same webContents.capturePage() as preview_snapshot, which fails with UnknownVizError on a hidden guest — so a tab opened with open: false, or one whose thread is not routed in the desktop window, will not produce an artifact.

Verification

Against a dev desktop build, an agent called preview_screenshot after preview_open and reported sizeBytes: 57265, matching the artifact written to disk byte for byte. The image below is that artifact, unmodified — full resolution, not downscaled the way snapshot captures are.

tsgo --noEmit is clean for contracts, server, and web; the preview toolkit and automation broker tests pass.

Model: Claude Opus 5, via Claude Code.

Note

Add preview_screenshot tool to let agents capture browser tab screenshots

• Adds a new preview_screenshot MCP tool to PreviewToolkit and PreviewStandardToolkit that captures a full-resolution PNG of a targeted browser tab and returns a PreviewAutomationScreenshotArtifact with path, size, and timestamp.
• Extends PreviewAutomationHost to handle the screenshot automation operation by calling bridge.captureScreenshot with the runtime tab id.
• Adds the screenshot operation to PREVIEW_AUTOMATION_OPERATIONS and defines the PreviewAutomationScreenshotArtifact schema in previewAutomation.ts.

^{Macroscope summarized 63b662c.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 1c8488f6-467d-431c-93d7-0e54285bb198

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}