What it does
Cornea builds an abstract geometry model of each element on a page, covering box, position, z-order and computed styles, and derives findings for overlap, overflow, WCAG AA contrast and an overall quality score from 0 to 1. The same input always produces byte-identical JSON. The checks are exposed as layout.* tools over stdio MCP, and the same engine is also available as a CLI and an HTTP API.
An agent building web pages can use it to find out whether a layout is broken and where, using a compact structured report instead of screenshots or raw DOM dumps. It can also be pointed at a URL, such as a running local dev server. In that mode it fetches the page, inlines external stylesheets and scripts, and inspects the result. A fidelity tool reports which CSS features are exact and which are approximated. Inline scripts can run against a minimal DOM shim before inspection.
The Rust build requires only cargo, with no browser, Node or system dependencies. The cornea command can also be installed from npm without a Rust toolchain.
Limitations and requirements
HTTPS pages cannot be captured because the binary has no TLS stack; a local http dev server or a separate fetch step is needed. URL capture is a snapshot in time. Async JavaScript APIs such as setTimeout and fetch are rejected and reported in js_notes. Full React or SPA mounting is experimental (Phase B). External scripts run only when capture has inlined them. Sources the engine does not apply, such as external stylesheets, media queries and unresolved colours, are listed in report warnings, and some CSS features are approximated rather than exact.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
Cornea, deterministic visual inspection
Deterministic visual inspection for AI agents. Overlap, overflow, contrast, quality.
v0.2.2As shown on registry.modelcontextprotocol.io when we last read it