IrisImage facts with pixel-level proof.

Dimensions, format, and metadata by default. Local Tesseract OCR only when you ask for it.

Quickstart

Tool reference

Geometry first

The fast profile, or no profile, returns dimensions, format, metadata, and trust warnings. It does not run OCR.

Text when named

quality or include_ocr runs local Tesseract. A missing binary is a gap, and the read still succeeds.

Pixels you can cite

Results keep the source path, hash, pixel bounds, route, warnings, and gaps. GPS is always redacted.

The difference

A screenshot is not a paragraph.
Measured pixels are.

Iris — image facts with pixel-level proof. Dimensions, format, and metadata by default; local Tesseract OCR only when requested. There is no generative vision model on this path.

What a guess says

A caption of the screenshot, with no width, no hash, and no box around the words.

What Iris returns

dimensions and format on every read · OCR lines {text, bbox, confidence} only when requested · a gap when Tesseract is missing.

One image. OCR stays off until you ask.

json

{ "path": "/absolute/path/to/screenshot.png" }

json

{
  "path": "/absolute/path/to/screenshot.png",
  "include_ocr": true,
  "ocr_languages": ["eng"]
}

The first call is fast. The second runs local Tesseract. include_ocr: false wins over profile: "quality". PNG bytes for a crop are returned only when include_region_image is true.

How it works

Three steps from a file to a citation

Step 1

Add it to your agent

One npx line. A stdio MCP server starts for Claude, Cursor, VS Code, Codex, or any other MCP client. No API key.

Step 2

Read geometry first

image_probe returns format, dimensions, and a hash, and it never runs OCR. read_image returns dimensions, format, and metadata until you set include_ocr or profile to quality.

Step 3

Cite the box

Use the path, the hash, and the pixel bounds. If OCR was not requested, or Tesseract is missing, do not invent the text.