Rendercheck

What rendercheck MCP server does

RenderCheck is designed for visual text assertions in browser tests. Instead of checking only the DOM, it examines a screenshot of the rendered page and uses OCR to determine which characters are visible to a user. A test can therefore detect text that is absent, incorrect, or too faint to read.

The documented package is installed as @rendercheck/sdk. Its examples cover Playwright assertions and the repository also provides Cypress guidance. The project is intended for local, offline processing: screenshots are read on the test machine, with no API key, upload, or network request required for OCR.

How it works

A browser test navigates to a page and invokes a text assertion against the rendered result. RenderCheck captures the page through the test framework, preprocesses the image when needed, and passes it to Tesseract OCR running locally. The recognized text is compared with the expected string.

Failures include the text that was expected, the OCR result, a confidence value, and a character-level difference when the values differ. Low-confidence recognition is returned and flagged rather than silently treated as correct. The repository notes that recognition quality depends on the image and does not provide a general accuracy percentage.

The repository also contains a separate browser-based image-to-text application. That application uses Tesseract.js and WebAssembly in the browser, supports image enhancement, and keeps image processing on the device.

Setup and configuration

For Playwright or Cypress test usage, install the development dependency with:

npm install --save-dev @rendercheck/sdk

The documented Playwright example imports test from @playwright/test and the assertion helper from @rendercheck/sdk/playwright. The repository requires Node.js 20 or newer and npm 10 or yarn 4 for development. Cypress support has been exercised with Cypress 13 on Electron under Linux only, so other Cypress environments should be treated as unverified.

The repository's browser application can be configured with VITE_UX_V2 to enable its newer interface, VITE_SITE_URL for absolute canonical and social URLs, and related Vite settings. These configure the web application, not a documented MCP server interface.

Tools and capabilities

The rendercheck MCP server materials describe these testing capabilities:

• Screenshot-based text assertions for rendered browser pages
• Local OCR through Tesseract.js and WebAssembly
• Playwright SDK support
• Cypress SDK support
• Confidence reporting and low-confidence flags
• Differences showing expected and recognized characters

The browser application additionally offers image preprocessing, per-word confidence scores, result copying and downloading, local history in its newer interface, keyboard navigation, and responsive themes.

Limitations and notes

The supplied repository materials do not specify MCP tools, an MCP protocol transport, an installation command for an MCP server, or client compatibility. They document RenderCheck as a browser OCR application and a test SDK instead. Consequently, developers evaluating an MCP deployment should verify whether a separate MCP wrapper exists before configuring an agent client.

OCR results depend on screenshot quality, layout, contrast, and the text itself. The project describes its benchmark suite as a small self-made set of examples rather than a general accuracy measure. Cypress coverage is limited to the stated Cypress 13, Electron, and Linux combination.