Iris MCP server

Iris

Image facts with pixel-level proof

Iris gives agents deterministic facts from screenshots, UI captures, forms, diagrams, charts, and other images. It returns dimensions, metadata, OCR regions, layout blocks, crops, and trust warnings without requiring a generative vision model.

npx -y @sylphx/iris

For Claude Code:

claude mcp add iris -- npx -y @sylphx/iris

The fastest useful workflow

{
  "path": "/absolute/path/to/screenshot.png",
  "include_ocr": true
}

The result includes image dimensions, hash, OCR lines with bounding boxes, layout blocks, a text map for agents, and explicit warnings or gaps.

Jobs Iris is built for

Tool surface

Predictable defaults

• fast returns deterministic metadata, geometry, and layout.
• quality explicitly enables OCR and local semantic helpers.
• Generative captions and object detection are opt-in and never authoritative.
• GPS metadata is redacted unless explicitly requested.
• Oversized or unsupported files fail with a structured error.

Why agents trust it

Every result keeps source path and hash, pixel locators, extraction route, confidence or warnings, and known gaps. OCR and optional model output never replace deterministic geometry.

Companion MCP tools

Each product is independent. Install only the tools your agent needs.

Development

bun install
bun run build
bun test
cargo test
bun run benchmark:public-proof
bun run benchmark:release-gate

License

MIT