AI-ready camera server bridging industrial automation and the Model Context Protocol (MCP)

camera-mcp is an experimental Java service from ErgoTech Systems that makes live camera feeds accessible to AI agents and LangChain4j / MCP clients — much like an OPC UA server exposes process variables to control systems.

Originally inspired by a small Python OPC UA prototype that published a camera frame as a ByteString tag specifically the excellent OPC-UA-Server-for-Camera project by RavilN this project re-implements the concept in Java using OpenCV and Spring AI MCP, adding:

• dynamic image encoding (JPEG/PNG),
• region-of-interest masking,
• change-based streaming,
• and configurable grayscale/color modes.

The goal: provide an AI-native vision node

flowchart LR
    subgraph OPC_UA["🧱 OPC UA Prototype (Python)"]
        A1["cv2.VideoCapture()"] --> A2[Percent difference & deadband]
        A2 --> A3["ByteString <br/> OPC UA tag \"CameraImage\""]
    end

    subgraph MCP["🌐 MCP Server (Java + LangChain4j)"]
        B1[OpenCV capture service] --> B2[MCP tools via Spring AI]
        B2 --> B3["SSE / JSON-RPC streaming <br/> (JPEG | PNG | Change)"]
        B3 --> B4[Agentic AI / MIStudio integration]
    end

    OPC_UA -->|Re-imagined for AI context| MCP

┌──────────────────────────┐
│  Camera (USB / IP)       │
└──────────┬───────────────┘
           │  OpenCV
           ▼
┌──────────────────────────┐
│  CameraService           │ → Frame capture & ROI masking
└──────────┬───────────────┘
           ▼
┌──────────────────────────┐
│  ChangeMetricsService    │ → Percent-difference computation
└──────────┬───────────────┘
           ▼
┌──────────────────────────┐
│  StreamController        │ → /stream/sse, /stream/sse-png, /stream/sse-png-change
└──────────┬───────────────┘
           ▼
┌──────────────────────────┐
│  MCP Server (Spring AI)  │ → /mcp JSON-RPC 2.0 endpoint
└──────────────────────────┘

• Java 21+
• Gradle 8.5+
• Webcam or IP camera accessible to OpenCV

./gradlew clean bootRun

Server starts (default port 8080) exposing:

Headless integration tests verify:

Run all tests:

./gradlew test

Example RPC call:

{
  "jsonrpc": "2.0",
  "id": "1",
  "method": "tools/call",
  "params": {
    "name": "get_image",
    "arguments": { "useGrayscale": true, "fmt": "png", "quality": 90 }
  }
}

Typical response:

{
  "jsonrpc": "2.0",
  "id": "1",
  "result": {
    "mimeType": "image/png",
    "data_base64": "<Base64 image>",
    "width": 640,
    "height": 480,
    "ts": 1697045290000
  }
}

Edit src/main/resources/application.yml:

server:
  port: 8080

spring:
  ai:
    mcp:
      server:
        http:
          enabled: true
          path: /mcp
          sse:
            enabled: true

Optional environment variables:

MCP brings contextual, agent-aware data virtualization to the same vision interfaces that OPC UA once standardized for automation.

• Embed as an MCP data source node
• Deploy anyewhere, eg on Jetson Nano / Raspberry Pi edge devices
• Feed live imagery into AI pipelines for anomaly detection
• Attach to Digital Twin dashboards via MCP streaming

• WebSocket / gRPC transport options
• Per-client configuration via MCP contexts
• Object detection / annotation overlays
• Multi-camera support
• Integration with ErgoTech MIStudio MCP mesh

ErgoTech Systems, Inc. (Los Alamos, NM) develops AI-driven manufacturing software including:

• MIStudio – a no/low-code platform for process monitoring and AI integration
• TransSECS – a configurable SECS/GEM interface builder for semiconductor tools
• Data Mesh & Digital Twin frameworks bridging OT, IT, and AI systems

Visit 🌐 ergotech.com for more information.

MIT License © 2025 ErgoTech Systems, Inc.

“MCP turns every data source into a live, streaming part of the world’s context layer — so AI, analytics, and automation all work from the same real-time truth.”
— ErgoTech Systems, Inc.