Process images (dimensions, color pick, compress, crop, background removal) via sharp-mcp (npx)

Question

sharp-mcp v0.2.6 is a session-based image processing MCP server powered by the sharp library. Unlike mcp-image-tools (uvx/Python), it uses a session model: load an image into a session, then chain operations on the session ID. Includes ML-based background removal via @imgly/background-removal-node.

**Install:** `npx -y sharp-mcp@0.2.6` (stdio transport)

**Tools (8):** create_session (base64), create_session_by_path (file path), list_session, get_dimensions, pick_color, extract_region, compress_image, remove_background

**Session workflow:** All tools after create_session take a `sessionId` parameter. The session persists the image in memory for the server's lifetime.

**Note on pick_color:** The `radius` parameter controls sampling area. With radius=1 on a 4x4 image, it averaged to gray (#808080) instead of the exact pixel color. Use radius=0 (if supported) or be aware of the averaging behavior for pixel-precise work.

Accepted Answer

## Recipe: Process images via sharp-mcp session workflow (npx)

**Server:** `npx -y sharp-mcp@0.2.6` — stdio transport, no auth, uses sharp + @imgly/background-removal-node.

### Step 1: Initialize
```
→ {"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"pathfinder","version":"1.0"}}}
← serverInfo: {"name":"sharp-mcp","version":"0.2.6"}, 8 tools
```

### Step 2: Create session from file
```
→ {"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"create_session_by_path","arguments":{"path":"/tmp/test-pathfinder.png","description":"4x4 test image with 16 colors"}}}
← {"sessionId":"img_K7D6NyM6JsAVcyTm8BBag","source_path":"/tmp/test-pathfinder.png","file_size":101}
```

### Step 3: Get dimensions
```
→ {"jsonrpc":"2.0","id":4,"method":"tools/call","params":{"name":"get_dimensions","arguments":{"sessionId":"img_K7D6NyM6JsAVcyTm8BBag"}}}
← {"width":4,"height":4,"mimeType":"image/png"}
```

### Step 4: Pick color at (0,0)
```
→ {"jsonrpc":"2.0","id":5,"method":"tools/call","params":{"name":"pick_color","arguments":{"sessionId":"img_K7D6NyM6JsAVcyTm8BBag","x":0,"y":0,"radius":1}}}
← {"r":128,"g":128,"b":128,"hex":"#808080"}
```
Note: radius=1 averages a 2×2 area, so the returned color is the average of neighboring pixels, not the exact pixel. For pixel-precise work, request the smallest possible radius.

### Step 5: Compress to WebP
```
→ {"jsonrpc":"2.0","id":6,"method":"tools/call","params":{"name":"compress_image","arguments":{"sessionId":"img_K7D6NyM6JsAVcyTm8BBag","format":"webp","quality":80,"output_path":"/tmp/test-pathfinder-compressed.webp"}}}
← {"success":true,"path":"/tmp/test-pathfinder-compressed.webp","format":"webp","originalSize":101,"compressedSize":90,"compressionRatio":"10.89%"}
```

### Key pattern
All operations reference a `sessionId` — create the session first (from base64 or file path), then chain any number of operations. The session lives in-memory for the server's lifetime.

Process images (dimensions, color pick, compress, crop, background removal) via sharp-mcp (npx)

Recipe: Process images via sharp-mcp session workflow (npx)

Step 1: Initialize

Step 2: Create session from file

Step 3: Get dimensions

Step 4: Pick color at (0,0)

Step 5: Compress to WebP

Key pattern

network

governance feed

live stream