process_image_to_description

Instructions

Implementation Reference

• src/index.ts:251-295 (handler)

  The main handler for the 'process_image_to_description' tool. It extracts image_url, detail_level, and focus parameters, downloads the image, constructs a descriptive prompt based on parameters, calls the processImageWithQwen helper to get the AI description, and returns the result as text content or error.

  case "process_image_to_description": {
    const imageUrl = String(request.params.arguments?.image_url);
    const detailLevel = String(request.params.arguments?.detail_level || "standard");
    const focus = String(request.params.arguments?.focus || "");
    
    if (!imageUrl) {
      throw new Error("图像URL是必需的");
    }
    
    try {
      // 下载图像
      const imagePath = await downloadImage(imageUrl);
      
      // 构建提示词
      let prompt = "请描述这张图片";
      if (detailLevel === "brief") {
        prompt += "，给出简洁的描述";
      } else if (detailLevel === "detailed") {
        prompt += "，提供详细的描述，包括细节、背景和上下文";
      }
      
      if (focus) {
        prompt += `，重点关注${focus}`;
      }
      
      // 处理图像
      const result = await processImageWithQwen(imagePath, prompt);
      
      return {
        content: [{
          type: "text",
          text: result
        }]
      };
    } catch (error: any) {
      return {
        content: [{
          type: "text",
          text: `处理图像失败: ${error.message}`
        }],
        isError: true
      };
    }
  }

• src/index.ts:181-201 (schema)

  Input schema definition for the tool, specifying parameters: image_url (required), detail_level (enum with default), focus (optional). Defines validation for tool arguments.

  inputSchema: {
    type: "object",
    properties: {
      image_url: {
        type: "string",
        description: "图像URL"
      },
      detail_level: {
        type: "string",
        description: "描述详细程度：简洁(brief)、标准(standard)或详细(detailed)",
        enum: ["brief", "standard", "detailed"],
        default: "standard"
      },
      focus: {
        type: "string",
        description: "描述重点，例如：物体、场景、人物、情感等",
        default: ""
      }
    },
    required: ["image_url"]
  }

• src/index.ts:178-202 (registration)

  Tool registration in the ListToolsRequestSchema handler, defining name, description, and input schema for discovery by MCP clients.

  {
    name: "process_image_to_description",
    description: "处理图像并生成描述",
    inputSchema: {
      type: "object",
      properties: {
        image_url: {
          type: "string",
          description: "图像URL"
        },
        detail_level: {
          type: "string",
          description: "描述详细程度：简洁(brief)、标准(standard)或详细(detailed)",
          enum: ["brief", "standard", "detailed"],
          default: "standard"
        },
        focus: {
          type: "string",
          description: "描述重点，例如：物体、场景、人物、情感等",
          default: ""
        }
      },
      required: ["image_url"]
    }
  }

• src/index.ts:74-131 (helper)

  Core helper function that calls the Qwen VL API: reads image to base64, sends prompt + image to API, extracts text response, cleans up temp file. Used by both image processing tools.

  async function processImageWithQwen(imagePath: string, prompt: string): Promise<string> {
    try {
      // 读取图像文件
      const imageBuffer = fs.readFileSync(imagePath);
      const base64Image = imageBuffer.toString('base64');
      
      // 准备请求数据
      const requestData = {
        model: "qwen-vl-plus",
        input: {
          messages: [
            {
              role: "user",
              content: [
                {
                  type: "text",
                  text: prompt
                },
                {
                  type: "image",
                  image: base64Image
                }
              ]
            }
          ]
        },
        parameters: {}
      };
      
      // 发送请求到API
      const response = await axios.post(API_ENDPOINT, requestData, {
        headers: {
          'Authorization': `Bearer ${API_KEY}`,
          'Content-Type': 'application/json'
        }
      });
      
      // 处理响应
      if (response.data && response.data.output && response.data.output.text) {
        return response.data.output.text;
      } else {
        throw new Error("API响应格式不正确");
      }
    } catch (error: any) {
      console.error("调用qwen2.5-vl模型API失败:", error);
      if (error.response) {
        console.error("API响应:", error.response.data);
      }
      throw new Error(`处理图像失败: ${error.message}`);
    } finally {
      // 清理临时文件
      try {
        fs.unlinkSync(imagePath);
      } catch (e) {
        console.error("清理临时文件失败:", e);
      }
    }
  }

• src/index.ts:47-66 (helper)

  Helper to download image from URL to temp file using stream, generates unique filename, returns path. Used in tool handlers.

  async function downloadImage(imageUrl: string): Promise<string> {
    try {
      // 为图像生成唯一文件名
      const filename = `${Date.now()}-${Math.random().toString(36).substring(2, 15)}.jpg`;
      const filePath = path.join(TEMP_DIR, filename);
      
      // 下载图像
      const response = await axios({
        method: 'GET',
        url: imageUrl,
        responseType: 'stream'
      });
      
      await streamPipeline(response.data, fs.createWriteStream(filePath));
      return filePath;
    } catch (error: any) {
      console.error("下载图像失败:", error);
      throw new Error(`下载图像失败: ${error.message}`);
    }
  }