# Iris

**Category:** 🎥 Multimedia Process  
**Repository:** https://github.com/SylphxAI/iris  
**GitHub Stars:** 2  
**npm Downloads (last month):** 23232  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/iris-2

## Description
Iris — image facts with pixel-level proof: metadata, OCR boxes, layout, and crops.

## Tools
Capabilities this server exposes over MCP:

- **read_image** — One local image. Geometry and metadata. OCR only when requested. An optional `region` crops inside this read.
- **image_probe** — Format, dimensions, pixel count, and source hash. No OCR and no crop.
- **crop_region** — One citeable region: bounds and hash. Set `include_region_image` for PNG bytes. No OCR.
- **compare_images** — Changed pixels between two images of equal dimensions. No OCR.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "iris": {
    "command": "npx",
    "args": ["-y","@sylphx/anymd"]
  }
}
```

## Documentation

## What the Iris MCP server does

The Iris MCP server gives an MCP client structured facts about local image files. It can report dimensions, format, pixel count, metadata, source hashes, and trust warnings without using a generative vision model. OCR is an opt-in operation, so a normal image read does not silently turn visual content into a transcription.

The server is suited to workflows where an agent must cite or verify image evidence. For example, it can identify the geometry of a screenshot, extract requested text with its location, isolate a region for reference, or measure changes between two captures. It does not provide general image understanding beyond the documented measurements and operations.

## How it works

Iris runs as a stdio MCP server from the npm package `@sylphx/iris`. A client sends local file paths and tool arguments, and the process returns structured results. No API key is required. OCR runs locally through the `tesseract` executable when requested; the default OCR language is `eng`.

`read_image` combines image geometry and metadata inspection, with an optional region and optional OCR. `image_probe` is narrower: it returns format, dimensions, pixel count, and a source hash, without cropping or OCR. `crop_region` produces the bounds and hash for one region and can include PNG bytes when requested. `compare_images` reports changed pixels and a changed bounding box for two images with matching dimensions.

## Setup and configuration

Run the server with:

```bash
npx -y @sylphx/iris
```

For Claude Code, the documented command is:

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

Other documented client configurations use `npx` as the command and `-y`, `@sylphx/iris` as its arguments. The repository lists Claude Desktop, Cursor, VS Code, and Codex with this configuration style. OCR requires the `tesseract` binary to be available on `PATH`; the server itself does not require an API key.

## Tools and capabilities

- `read_image`: Reads one local image, returning geometry and metadata. OCR can be enabled with `include_ocr: true` or `profile: "quality"`; an explicit `include_ocr: false` takes precedence.
- `image_probe`: Returns format, dimensions, pixel count, and source hash without OCR or cropping.
- `crop_region`: Returns a citeable region’s bounds and hash, with optional PNG region data.
- `compare_images`: Finds changed pixels between equal-sized images without resizing either input.

OCR results contain text, bounding boxes, and confidence values. The response can identify an unavailable OCR dependency through the `OCR_UNAVAILABLE` gap rather than fabricating text.

## Limitations and notes

The default file-size limit is 32 MiB. A cropped region cannot exceed 67,108,864 pixels. Files that are too large or cannot be decoded fail instead of being returned as successful reads. GPS metadata is always redacted, and the material does not describe an option to restore it.

`compare_images` requires identical dimensions and performs no resizing. OCR is not automatic, and missing Tesseract means OCR is unavailable even though non-OCR image inspection can remain successful. The Iris MCP server repository was archived on September 25, 2026 after its functionality was merged into anymd, which also reads other document and media formats. This listing therefore describes a documented package and interface from an archived repository, not an actively developed codebase.

_Full upstream README: https://allmcps.com/mcp/iris-2/readme_