OceanGPT·Cangyuan has newly launched the Sonar Image MCP (Model Context Protocol) Service, which supports the dynamic invocation of Cangyuan Large Model's sonar image interpretation capabilities via MCP. MCP is a universal model context extension protocol, whose core function is to provide a pluggable professional service invocation mechanism for large models, enabling seamless integration of external capabilities through context protocol interfaces. In sonar image scenarios, MCP allows users to embed the sonar image analysis capabilities of the Cangyuan Model into their own large-model applications, converting raw sonar images into natural language descriptions and supporting integrated reasoning of "sonar images + text". This capability is applicable to application scenarios such as underwater environment modeling, target detection, and mission planning, enabling researchers and engineering applications to directly obtain underwater perception results through a unified conversational agent interface.
This document will introduce the Sonar Image Intelligent Recognition and Perception MCP Service and its usage from the following aspects:
MCP (Model Control Protocol) is a protocol for large-model API invocation and management. It can serve as an intermediate layer between AI and various external tool services, replacing humans to access and operate external tools, and greatly simplifying the complexity of API invocation. Based on this Sonar Image Intelligent Recognition and Perception MCP Service, users can interact with the large model in a conversational manner to achieve sonar image perception and natural language interpretation.
MCP Service Address: http://60.191.72.39:8082/sse
MCP Service User Documentation: https://github.com/OceanGPT/OceanGPT/blob/main/mcp_server/README_CN.md
This service is built on a dedicated model trained with public sonar image datasets and currently supports perception capabilities for the following 15 types of objects:
| No. | Object Name (Chinese) | Object Name (English) |
|---|---|---|
| 0 | 球 | Ball |
| 1 | 圆形笼 | Circle Cage |
| 2 | 立方体 | Cube |
| 3 | 圆柱体 | Cylinder |
| 4 | 瓶状物 | Bottle |
| 5 | 金属桶 | Metal Bucket |
| 6 | 飞机 | Plane |
| 7 | 水下机器人 | ROV (Remotely Operated Vehicle) |
| 8 | 方形笼 | Square Cage |
| 9 | 轮胎 | Tyre |
| 10 | 船体 | Ship (Hull) |
| 11 | 钩子 | Hook |
| 12 | 螺旋桨 | Propeller |
| 13 | 气阀 | Valve |
| 14 | 水下管道 | Underwater Pipeline |
The following sections will introduce how to use this MCP Service based on Cherry Studio and Cursor.
For MCP-related usage methods, please refer to the Bilibili video: https://www.bilibili.com/video/BV1RNTtzMENj
To facilitate testing, we provide the following two methods to obtain test image URLs: