kelra.mcp

A visual reasoning tool for coding agents

An interface is more than what you see.

Kelra reads a reference screen and gives your coding agent evidence it can build from: what’s visible, how it may fit together, and what the image can’t confirm.

Follow the evidence

Early development · Screenshot analysis is implemented

Kelra’s colorful spectrum mark

From a screenshot to a plan an agent can reason about.

Reference in → Evidence out

A useful first pass

Keep the clues.
Keep the uncertainty.

A screenshot contains more than pixels to copy. Kelra separates what can be seen from what must be reasoned about.

REFERENCE SCREENSTRUCTURED EVIDENCE
reference.png
visible textsurface edge
ObservedHeading and button text
from image
InferredCards share a grid
estimated
UncertainExact spacing values
not visible
Evidence stays attached to the spec

Illustrative example · The actual output depends on the reference screen

A clear handoff

The agent gets a guide.
You keep the work.

Kelra produces a platform-neutral reconstruction spec. Your coding agent still writes the app, runs it, makes corrections, and works with Git.

MCP transport and code generation are still in development.

A visual reasoning step in your workflow
Reference
screen
Kelra
visual evidence
Coding
agent

Understanding the reference · implementation remains yours

Kelra MCP

Build from what’s there.

Screenshot analysis for agents that build interfaces. In active development.