Locators for your coding agent,
graded before they ship.
Your agent opens a live page and gets ranked, graded locators in the test framework your repo already uses.
Add it to your client once. Your agent runs it on demand.
Nothing is installed globally. Your agent runs the server with npx.
To share it with your team, put this in .mcp.json at the project root:
Other clients
Any other MCP client: register a stdio server whose command is npx with arguments -y @corlabsdev/mcp.
Eight tools, one loop: open, inspect, place, check.
Node 18+ and a Chrome-family browser.
Google Chrome works. Microsoft Edge or Playwright’s Chromium also work. If no browser is found, the server says so. Install Chrome, or run npx playwright-core install chromium.
Questions
Does it send data off my machine?
Yes, anonymous usage metrics, on by default in the published package. Two events: mcp_installed once per install (package version, Node version, platform), and mcp_tool_called per tool call (tool name, duration, outcome, package version). Each install has a random ID stored in ~/.cor/install-id. Page URLs, element text, locators, file paths, input values, cookies, error messages and email addresses are never sent. Turn it off with COR_TELEMETRY=off, or set COR_OFFLINE=true to stop all outbound data.
Where are the locators stored?
In your project, under .cor/records/. Placed entries are meant to be committed. Entries shown but not yet placed are git-ignored, so they never reach a commit.
Can it run without a visible browser?
Yes. Pass --headless after the package name. The default opens a visible window.
Which test frameworks does it write for?
The package is built for Playwright, Cypress, Selenium and WebdriverIO. It writes locators in the framework your repo already uses.