screenroveSCREENS WORTH STUDYING
For agents

Give your agent real screens to build from.

Screenrove is an MCP server: free, no account or API key. Connect it once and your coding agent can search 6,950 captured screens from Linear, Cursor, Vercel, Airbnb, Stripe, Duolingo, Notion, Raycast and 183 more products, show you a few to choose from, and build from the one you pick. No one there to pick? It works on its own too.

Connect

Claude

claude.ai, Claude Desktop and the mobile apps
Add to Claude

Opens claude.ai with Screenrove filled in as a custom connector. If the form comes up empty, open Settings → Connectors → Add custom connector and paste this URL:

https://screenrove.com/mcp

Works on every plan; Free allows one custom connector. On Team and Enterprise an owner has to allow custom connectors first. Checked 25 October 2026.

On a phone? Adding a connector needs a one-time sign-in to claude.ai in your phone’s browser; the app’s sign-in doesn’t carry over. After that, Screenrove appears in the Claude app too. It’s easier on a computer: send yourself this page.

No connector? This prompt works in any chat. Paste it into Claude and fill in your idea.

Check it worked: start a new chat, open + → Connectors and check Screenrove is on.

Then ask: “Find me pricing page references on Screenrove and let me pick one.”

Claude Code

Terminal
claude mcp add --transport http --scope user screenrove https://screenrove.com/mcp

Run it in your terminal (Terminal, iTerm or PowerShell), not inside a Claude Code session. Or paste it into Claude Code and ask it to run it. --scope user makes Screenrove available in every project.

Check it worked: start a new session with claude and type /mcp: screenrove should say connected.

Then ask: “Find me three onboarding flows on Screenrove and let me pick one for my signup.”

Cursor

One click, or one entry in mcp.json
Add to Cursor

Opens Cursor and asks to install. If nothing happens, Cursor isn’t installed; or add it by hand. Open mcp.json in the hidden .cursor folder in your home folder (macOS and Linux: ~/.cursor/mcp.json; Windows: %USERPROFILE%\.cursor\mcp.json) and add this entry inside "mcpServers", after a comma if other servers are already there:

"screenrove": {"url": "https://screenrove.com/mcp"}

No file yet? Create it with this:

{"mcpServers": {"screenrove": {"url": "https://screenrove.com/mcp"}}}

Check it worked: open Cursor Settings → MCP (Tools & MCP in newer versions): screenrove should show a green dot and its tools. Restart Cursor if it doesn’t.

Then ask: “Use Screenrove to find empty state examples with an illustration, then build ours.”

VS Code

GitHub Copilot agent mode
Add to VS Code
code --add-mcp '{"name":"screenrove","type":"http","url":"https://screenrove.com/mcp"}'

Or by hand: add this entry inside "servers" in .vscode/mcp.json (this project) or your user configuration (Command Palette → MCP: Open User Configuration):

"screenrove": {"type": "http", "url": "https://screenrove.com/mcp"}

Then trust and start the server when VS Code asks (or Command Palette → MCP: List Servers → screenrove → Start Server), switch Copilot Chat to Agent and turn on Screenrove in the tools picker.

Check it worked: MCP: List Servers shows screenrove as running, and the tools picker lists its tools.

Then ask: “Use Screenrove to show me real settings pages, then build our settings screen like the one I pick.”

Codex

CLI and IDE extension
codex mcp add screenrove --url https://screenrove.com/mcp

Run it in your terminal, then start a new Codex session.

Check it worked: run codex mcp list: screenrove should be listed.

Then ask: “Get Linear’s design guide from Screenrove and use its product palette for our settings screen.”

Windsurf / Devin

Devin Desktop and CLI
devin mcp add -s user screenrove https://screenrove.com/mcp

Cascade: add this entry inside "mcpServers" in ~/.config/devin/mcp_config.json:

"screenrove": {"serverUrl": "https://screenrove.com/mcp"}

Check it worked: Cascade’s list of MCP servers shows screenrove with its tools.

Then ask: “Show me three paywalls with a free trial on Screenrove and let me pick one.”

ChatGPT

chatgpt.com, Business, Enterprise and Edu workspaces

Checked 25 October 2026: OpenAI’s help centre lists custom MCP servers (developer mode, in beta) for ChatGPT Business, Enterprise and Edu on chatgpt.com, and a workspace owner or admin has to turn developer mode on first. Free, Go, Plus and Pro aren’t on that list, so you may not see the option.

  1. Open chatgpt.com/plugins in a browser on your computer.
  2. Press +, then Add custom MCP server.
  3. Name it Screenrove, paste this URL, choose no authentication, confirm the warning and create it.
https://screenrove.com/mcp

Can’t add it on your plan? No setup needed: paste a research prompt with image links into any chat. This one compares the plan sections of Linear, Notion and monday.com. For your own picks, save or compare screens, then choose Use these in agent → I’m researching, not building.

Check it worked: in a new chat, turn Screenrove on from the + menu and ask which Screenrove tools it can see.

Then ask: “Use Screenrove to find three real pricing pages and compare them in a table with capture dates.”

Anything else

Any MCP client
https://screenrove.com/mcp

Streamable HTTP. No key or account.

Looking for more MCP servers for Claude Code? Claude Code Agents, our directory of MCP servers, lists thousands.

Check it worked: your client lists 24 Screenrove tools, from build_handoff to wait_for_pick.

Then ask: “Search Screenrove for login pages with a magic link and show me the best three.”

How it works

  1. Ask for a reference. “Find me some pricing page references and let me pick one.”
  2. Pick in your browser. Your agent opens a page with a few real screens side by side. Hover to scroll a long page, click the one you want. In Claude and ChatGPT the picker can appear right in the chat.
  3. Build from it. Your agent gets the screenshot, notes on what the screen does, the measured design guide and a build prompt.

Autonomous agents

No one there to pick? Screenrove works without a person in the loop.

  1. Search. search_screens with tag filters (list_tags gives the vocabulary), or find_related_screens from a screen you already have.
  2. Look before you build. Before writing UI, call build_recipe with your 2–6 picks or get_screen_image on your top picks. Builds from the text notes alone come out generic.
  3. Cite what you used. Name each screen id and its capture date in your PR or report. Every screen is observed on that date, and the product may have changed since.

Skip pick_reference and wait_for_pick: they wait for a person to choose.

Try asking

Tools

build_handoffBuild an agent handoff prompt
build_recipeBuild a reference recipe
check_my_uiCheck the user's UI against real references
compare_pricingCompare pricing pages as a fact table
compare_screensCompare two screens
find_related_screensFind related screens in other products
get_designGet a product's design guide
get_flowGet a product's onboarding flow
get_motionGet one animation with its spec and code
get_motion_guidelinesGet motion guidelines and easing tokens
get_productGet one product
get_screenGet one screen
get_screen_imageGet a screen's screenshot
list_flowsList captured onboarding flows
list_productsList captured products
list_tagsList design-problem tags
name_patternsGet naming pattern statistics
pick_referenceLet the user pick a reference
search_logosSearch the logo directory
search_motionSearch UI animations and micro-interactions
search_namesSearch real product names and their naming patterns
search_screensSearch screens
suggest_polishSuggest motion and sound polish for a UI
wait_for_pickWait for the user's pick

Fewer tools

Your client loads all 24 tools by default. To load only the ones you need, add tools to the server URL: a profile (screens, names, logos, motion or all) or tool names, separated by commas.

https://screenrove.com/mcp?tools=screens
https://screenrove.com/mcp?tools=names,logos

Prefer HTTP?

Everything is also a REST API. See the OpenAPI spec or llms.txt. Usage is recorded anonymously, as described in our privacy notes.