UI Verify MCP server: visual testing for Claude Code, Cursor and Codex

UI Verify is visual regression testing for coding agents, for Storybook, Playwright, and Vitest, and its MCP server lets the agent that wrote a UI change review it in the same conversation. A screenshot only shows an agent the page as it is now; UI Verify crops each change to what moved, so the agent looks at the difference instead of scanning a full page for it. The endpoint is https://uiverify.ai/mcp, and you log in once with your UI Verify account in the browser, so the config holds no API key.

Run claude mcp add --scope project --transport http uiverify https://uiverify.ai/mcp in your project root: it writes the server to your committed .mcp.json, so the whole team gets it. Then restart Claude Code, run /mcp, pick uiverify, and log in to UI Verify in your browser.

Claude Code keeps the login and renews it, so you do not log in again every session.

Add a uiverify entry with the URL https://uiverify.ai/mcp under mcpServers in .cursor/mcp.json, then log in to UI Verify from Cursor's MCP settings.

Run codex mcp add uiverify --url https://uiverify.ai/mcp: Codex adds the server to its config and opens the UI Verify login in your browser right away. To log in again later, run codex mcp login uiverify.

Any other MCP client takes the same URL and asks you to log in. In Claude.ai and Claude Desktop, connect UI Verify from Claude's connector directory. The config holds no secret, so you can commit it and each teammate logs in with their own account. Restart the agent session after adding the server so its tools load.

• Find the build for its commit or pull request, and see which stories changed, which failed, and which matched their baseline.
• Look at the change itself: render_diff_image returns a before and after cropped to what moved, as pixels its vision model reads directly, not a URL to download.
• Accept an intended change as the new baseline, deny a regression, or accept a whole build at once.
• Read the review comments teammates left on a diff, address them, and resolve the thread.
• See what a whole pull request changed in the UI against its base branch, including new and removed stories.

The agent does not have to wait for a pull request: `uiverify check` renders just the stories an edit touched, and the agent reviews that preview build the same way. The MCP tool reference lists every tool and its arguments, and Triage visual changes from your coding agent walks through the review.

Reviewing a change is an explicit call: the agent accepts or denies it for you and is recorded as doing so, and you can make the same call in the dashboard. A CI build you run with --auto-accept-changes, usually on your default branch, accepts its changes without a review (see baselines). Otherwise the UI Verify check does not pass until each change is reviewed, and a denied change fails it; make it a required check and nothing merges past it. With flake detection on (the default), a diff that does not reproduce when UI Verify re-renders the story comes back marked unstable and is left out of the check, so the agent can tell a flake from a real change (see flake detection).

Playwright MCP gives the agent a browser, and the UI Verify MCP server gives it the diff against an approved baseline. Playwright MCP, the Playwright team's server, lets an agent navigate, click, and fill forms on a running app, and read the page through its accessibility snapshot or a screenshot. Its README describes no baseline and no screenshot diff (checked 2026-10-25), so the agent sees the page as it is now and judges by itself whether anything regressed. UI Verify compares every capture against the approved baseline and returns the stories that changed, each with its diff.

The two fit together: Playwright MCP while the agent builds and pokes at the app, UI Verify when the change has to be checked before it merges. For the other visual testing MCP servers, each with its own sources, see visual testing MCP servers, compared.

The UI Verify MCP server works with every UI Verify build, whatever produced it: Storybook stories, Playwright tests (swap the @playwright/test import for @uiverify/playwright), Vitest browser mode tests (add the @uiverify/vitest plugin), and custom screenshots from any other tool. The server reads the build, not your test runner, so the agent reviews all of them the same way.

The UI Verify MCP server comes with every plan, the free one included. The free tier is 10,000 snapshots a month, and 50,000 for open-source projects, self-declared at sign-up with no card (see free for open source). Reviewing over MCP adds nothing to the bill: what counts is the snapshots your builds use (see how snapshots are billed). See pricing for the paid plans.