Read, resize, crop, and convert images locally via mcp-image-tools (uvx)

Recipe: inspect, resize, and convert images locally via mcp-image-tools

Surface

• Package: mcp-image-tools (PyPI)
• Version: 1.0.0 (server reports as MCP Server v1.27.2, built on mcp-server-framework)
• Transport: stdio
• Launch: uvx mcp-image-tools
• Auth: none (fully local, Pillow-based, credential-free)
• Tools: 6 — image_info, image_read_base64, image_resize, image_crop, image_screenshot, image_convert

What it does

Gives agents local image manipulation via Pillow — no external APIs, no network. Common agent use cases:

• Get metadata: dimensions, format, color mode, file size — without loading the full image into context
• Resize: generate thumbnails or fit images to constraints, returned as base64 PNG
• Crop: extract a specific region from an image
• Convert: change format (PNG→JPEG, JPEG→WebP, etc.) with configurable quality
• Read: load any image as base64 for passing to multimodal models

Verified trace (2 tool calls, both succeeded)

1. Get image metadata:

→ tools/call  image_info  {path: "/tmp/test-image.png"}
← {
    "path": "/private/tmp/test-image.png",
    "format": "PNG",
    "mode": "RGB",
    "width": 800,
    "height": 600,
    "size_bytes": 2791
  }
  isError: false

2. Resize image to 400×300:

→ tools/call  image_resize  {path: "/tmp/test-image.png", width: 400, height: 300}
← {result: "iVBORw0KGgoAAAANSUhEUgAAAZAAAAEs..."} (base64-encoded 400×300 PNG)
  isError: false

MCP config

{
  "mcpServers": {
    "image": {
      "command": "uvx",
      "args": ["mcp-image-tools"]
    }
  }
}

Tool reference

Failure modes observed

• None for file-based tools. Both image_info and image_resize returned correctly.
• image_screenshot likely requires a display server — will fail in headless/CI environments (not tested).
• All image results are base64-encoded, so large images produce large text responses. Use max_width on image_read_base64 to limit output size.

When to reach for this

• Agent needs to check image dimensions before uploading (e.g., "is this image at least 1200px wide?")
• Generating thumbnails or resized versions as part of a content pipeline
• Converting between formats (e.g., PNG→WebP for web optimization)
• Cropping a specific region from a screenshot or diagram for analysis
• Reading an image as base64 to pass to a multimodal model API