AI-Ready Screenshots
Let coding agents find your screenshots by what is on them and drop always-current embed URLs straight into your docs.
The problem
Agents that write docs and release notes cannot see your screenshot library, so they guess file names, link stale images, or ask you to go find the right one.
How ScreenshotAgent solves it
Every capture records what was on screen (page title, headings, a short description, UI element labels and visible text) into a searchable index for your team. Coding agents query it through the ScreenshotAgent MCP server and get back screenshot ids and public embed URLs.
Example: Claude Code finds your onboarding screenshots
Connect the ScreenshotAgent MCP server to Claude Code with a read-scoped team API token, then ask in plain language:
Public screenshots return an always-latest image URL and a ready-made Markdown snippet. Private screenshots are found too, but never get a public embed URL.
> Find the onboarding screenshots and give me their embed URLs.
- 1.Claude Code calls search_screenshots with a query like "onboarding welcome screen".
- 2.ScreenshotAgent searches screenshot names, tags, projects, URLs and what each capture saw on screen, by meaning and by keyword, best match first.
- 3.Each result comes back with its id, a snippet, whether it matched by keyword or by meaning, the page title and description, and public embed URLs.
- 4.Claude Code can call get_screenshot for full detail or recapture to refresh a page before using it.
Expected outcomes
- Ask for "the billing settings page in dark mode" and get the matching screenshot back.
- Agents paste stable, always-latest embed URLs instead of uploading one-off image files.
- The same search works in the app, over the REST API, and through MCP tools for Claude Code and other clients.
FAQ
- Is this semantic search?
- Yes. Search matches by meaning, not just keywords: each screenshot's recorded text and description are embedded with a text-embedding model and compared with the query, and those matches are merged with full-text keyword results. "change password" finds a screen labelled "Update credentials". Quoted phrases and -exclusions still match exactly. It searches the text and descriptions recorded at capture time, not the image pixels.
- Which AI tools can use it?
- Claude Code, Cursor and any other MCP client that supports the Streamable HTTP transport, through the search_screenshots tool. The same search is available at GET /api/v1/screenshots/search.
- Does it store passwords or secrets from my pages?
- The capture agent is told never to record passwords, API keys or tokens, and ScreenshotAgent redacts anything that looks like a key or credential before storing the metadata. Search is always scoped to your team.
- Do existing screenshots become searchable?
- Yes. Existing screenshots are searchable by name, tags, project and URL right away, by keyword and by meaning; on-screen details and the description of the layout are added the next time each page is captured.
Related features
- AI AnnotationsCircle, highlight, or point at the part of a screenshot that matters. AI finds it, we draw on top, and your screenshot's pixels are never changed.
- Searchable Screenshot LibraryEvery screen of your product that you have captured, in one library the whole team can search by meaning, not just keywords. Find a screen by what it shows, kept current on a schedule or after each deploy, with no extra cost per seat.
- Automated ScreenshotsCapture product screenshots automatically so docs and marketing visuals stay current.
Automate this workflow in your team
Start a free trial and connect your first project to see ScreenshotAgent capture and keep pages fresh.