grade_homework

Instructions

Implementation Reference

• src/tools/gradingTools.ts:46-116 (handler)

  Main handler for the 'grade_homework' tool call. Validates input (Base64 image or URL), downloads/processes image, invokes grading service, and formats the MCP response.

    async handleGradeHomework(args: any): Promise<any> {
      try {
        // 验证输入参数 - 支持Base64和URL
        const schema = z.union([
          z.object({ imageData: z.string().min(1, '图片数据不能为空') }),
          z.object({ imageUrl: z.string().url('请输入有效的图片URL地址') })
        ]);
  
        const validatedParams = schema.parse(args);
  
        // 获取图片数据
        let imageData: string;
        
        if ('imageUrl' in validatedParams) {
          imageData = await downloadImageAsBase64(validatedParams.imageUrl);
        } else {
          const validation = validateBase64Image(validatedParams.imageData);
          if (!validation.isValid) {
            throw new Error(validation.error || '图片格式验证失败');
          }
          imageData = validatedParams.imageData;
        }
  
        // 执行批改
        const result = await this.gradingService.gradeHomework({
          imageData: imageData,
          subject: '自动识别',
          studentName: '学生'
        });
        
        // 构建统一格式的题目输出
        const questionsOutput = result.results.map((r, i) => {
          const status = r.isCorrect ? '正确' : '错误';
          const questionContent = r.questionContent || `第${i + 1}题`;
          return `题号：${i + 1}
  题目：${questionContent}
  答案：${r.studentAnswer} （${status}）
  题目解析：${r.explanation}`;
        }).join('\n\n');
        
        return {
          content: [
            {
              type: 'text',
              text: `✅ 作业批改完成！
  
  📊 批改结果：
  • 总分：${result.totalScore}/${result.maxTotalScore}
  • 等级：${result.grade}
  
  📝 题目详情：
  ${questionsOutput}
  
  💭 总体评价：
  ${result.overallFeedback}`,
            },
          ],
        };
  
      } catch (error) {
        return {
          content: [
            {
              type: 'text',
              text: `❌ 作业批改失败：${error instanceof Error ? error.message : '未知错误'}`,
            },
          ],
          isError: true,
        };
      }
    }

• src/index.ts:47-49 (registration)

  Registers 'grade_homework' tool in the MCP ListTools response.

  return {
    tools: [gradeHomeworkTool],
  };

• src/index.ts:61-63 (registration)

  Routes 'grade_homework' tool calls to the ToolHandler in the MCP CallTool request handler.

  case 'grade_homework':
    return await this.toolHandler.handleGradeHomework(args);

• src/tools/gradingTools.ts:11-31 (schema)

  MCP tool definition for 'grade_homework', including name, description, and input schema.

  export const gradeHomeworkTool: Tool = {
    name: 'grade_homework',
    description: '📝 智能批改学生作业图片，支持Base64和URL两种方式，自动识别题目并给出评分和解析',
    inputSchema: {
      type: 'object',
      properties: {
        imageData: {
          type: 'string',
          description: 'Base64编码的作业图片数据（支持PNG、JPG、JPEG格式），与imageUrl二选一',
        },
        imageUrl: {
          type: 'string',
          description: '作业图片的URL地址，与imageData二选一',
        },
      },
      oneOf: [
        { required: ['imageData'] },
        { required: ['imageUrl'] }
      ],
    },
  };

• src/services/modelService.ts:25-100 (helper)

  Core helper function that invokes the vision-language model (Qwen-VL via OpenAI API) to analyze the homework image and produce grading results.

    async gradeHomework(
      imageData: string,
      subject: string,
      studentName: string,
      questions?: Question[]
    ): Promise<GradingResult[]> {
      try {
        logger.info(`开始批改${studentName}的${subject}作业`);
  
        // 准备图片数据
        const imageBase64 = convertToOpenAIImageFormat(imageData);
  
        // 构建提示词
        const prompt = this.buildGradingPrompt(subject, studentName, questions);
  
        // 调用模型
        const response = await this.client.chat.completions.create({
          model: CONFIG.model.model,
          messages: [
            {
              role: 'system',
              content: `你是一个专业的作业批改老师。请仔细查看学生提交的作业图片，逐题批改并给出详细的评分和反馈。
  
  要求：
  1. 准确识别图片中的题目和学生答案
  2. 逐题判断答案正确与否
  3. 为每道题提供简洁明了的解析说明
  4. 给出具体的得分和满分
  5. 提供建设性的反馈意见
  6. 输出格式必须为JSON格式
  
  评分标准：
  - 完全正确：满分
  - 部分正确：给部分分数
  - 完全错误：0分
  - 步骤正确但答案错误：给步骤分` 
            },
            {
              role: 'user',
              content: [
                {
                  type: 'text',
                  text: prompt,
                },
                {
                  type: 'image_url',
                  image_url: {
                    url: `data:image/jpeg;base64,${imageBase64}`,
                  },
                },
              ],
            },
          ],
          max_tokens: CONFIG.model.maxTokens,
          temperature: CONFIG.model.temperature,
          stream: false,
        });
  
        const content = response.choices[0]?.message?.content;
        if (!content) {
          throw new Error('模型返回内容为空');
        }
  
        logger.debug('模型返回内容:', content);
  
        // 解析模型返回的JSON结果
        const gradingResults = this.parseModelResponse(content);
        
        logger.info(`作业批改完成，共批改 ${gradingResults.length} 道题`);
        return gradingResults;
  
      } catch (error) {
        logger.error('调用模型批改作业失败:', error);
        throw new Error(`模型调用失败: ${error instanceof Error ? error.message : '未知错误'}`);
      }
    }