# tan-yong-sheng/ai-vision-mcp [Health: Active]

**Category:** 🎥 Multimedia Process  
**Repository:** https://github.com/tan-yong-sheng/ai-vision-mcp  
**GitHub Stars:** 80  
**npm Downloads (last month):** 195  
**Views:** 6  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/tan-yong-sheng-ai-vision-mcp

## Description
Multimodal AI vision MCP server for image, video, and object detection analysis. Enables UI/UX evaluation, visual regression testing, and interface understanding using Google Gemini and Vertex AI.

## Tools
Capabilities this server exposes over MCP:

- **analyze_image** — Analyze an image using AI vision models. Supports URLs, base64 data, and local file paths.
- **compare_images** — Compare multiple images using AI vision models. Supports URLs, base64 data, and local file paths.
- **detect_objects_in_image** — Detect objects in an image using AI vision models and generate annotated images with bounding boxes. Supports URLs, base64 data, and local file paths. File handling: explicit filePath → exact path, otherwise → temp directory. Uses optimized default parameters for object detection.
- **analyze_video** — Analyze a video using AI vision models. Supports URLs and local file paths.

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

```json
"mcpServers": {
  "ai-vision-mcp": {
    "command": "npx",
    "args": ["-y","ai-vision-mcp"],
    "env": {
      "IMAGE_PROVIDER": "",
      "VIDEO_PROVIDER": "",
      "GEMINI_API_KEY": "",
      "VERTEX_CLIENT_EMAIL": "",
      "VERTEX_PRIVATE_KEY": "",
      "VERTEX_PROJECT_ID": "",
      "GCS_BUCKET_NAME": ""
    }
  }
}
```

**Requires environment variables:** `IMAGE_PROVIDER`, `VIDEO_PROVIDER`, `GEMINI_API_KEY`, `VERTEX_CLIENT_EMAIL`, `VERTEX_PRIVATE_KEY`, `VERTEX_PROJECT_ID`, `GCS_BUCKET_NAME` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What tan-yong-sheng/ai-vision-mcp MCP server does

tan-yong-sheng/ai-vision-mcp MCP server brings multimodal image and video analysis into MCP-compatible clients. It connects to either the Google AI Studio Gemini API or Google Vertex AI, selected separately for image and video processing through environment variables.

The server is suited to interface review and visual inspection tasks. It can describe image content, compare multiple images, extract design tokens, assess visual hierarchy and eye flow, and inventory UI components while considering design-system maturity. These capabilities can support screenshot review, visual regression analysis, and UI/UX evaluation.

## How it works

The server runs over standard stdio transport and can be started with `npx ai-vision-mcp`. An MCP client sends a tool request with one or more image sources, a natural-language prompt, and optional analysis settings. Sources may be remote URLs, local file paths, or base64-encoded data.

Image analysis accepts one source and supports general analysis plus `palette`, `hierarchy`, and `components` modes. The implementation uses Zod validation, retry handling, and circuit-breaker behavior, according to the project documentation.

## Setup and configuration

Choose Google AI Studio by setting both provider variables to `google` and providing `GEMINI_API_KEY`. For Vertex AI, set both provider variables to `vertex_ai` and provide service-account details through `VERTEX_CLIENT_EMAIL`, `VERTEX_PRIVATE_KEY`, and `VERTEX_PROJECT_ID`. Vertex AI setup also requires `GCS_BUCKET_NAME` for Google Cloud Storage.

The README provides configurations for Claude Desktop, Claude Code, Cursor, and Cline. Other MCP-compatible clients can use the same stdio command. The examples recommend allowing more than five minutes for tool execution. Claude Code users are specifically advised to set a 60-second startup timeout and a 300-second tool timeout.

## Tools and capabilities

The documented tools include:

- `analyze_image`: Sends an image and prompt for general analysis or a selected design-focused mode.
- `compare_images`: Compares two to four images from URLs, local paths, or base64 data.
- `palette`: Extracts colors, spacing, and typography as design-token information through image analysis.
- `hierarchy`: Examines visual hierarchy and the likely eye-flow path through an interface.
- `components`: Lists visible UI components and evaluates design-system maturity.

Optional analysis settings include temperature and maximum output tokens.

## Limitations and notes

The server depends on access to Google Gemini or Vertex AI and requires the matching credentials before analysis can run. Vertex AI additionally depends on a Google Cloud Storage bucket. Image comparison is limited to a minimum of two and a maximum of four sources. Processing time can be substantial, so MCP client timeout settings may need adjustment.

_Full upstream README: https://allmcps.com/mcp/tan-yong-sheng-ai-vision-mcp/readme_