Code Review MCP Server

Instructions

Implementation Reference

• src/index.ts:71-128 (handler)

  The main handler function for the 'code_review' MCP tool. It executes repomix to flatten the repository, creates a CodeReviewService instance, calls reviewCodeFromRepomix to perform the LLM-powered review, handles errors, and returns structured JSON results.

  async (params) => {
    try {
      // Execute Repomix to get the flattened codebase
      const repomixOptions: RepomixOptions = {
        includePaths: params.specificFiles || [params.repoPath],
        fileTypes: params.fileTypes,
        outputFormat: 'plain',
      };
      
      const repomixOutput = await executeRepomix(repomixOptions);
      
      // Set up review options
      const reviewOptions: CodeReviewOptions = {
        detailLevel: params.detailLevel || 'detailed',
        focusAreas: params.focusAreas || ['security', 'performance', 'quality', 'maintainability'],
      };
      
      // Create the code review service
      try {
        const codeReviewService = createCodeReviewService();
        
        // Perform the code review
        const reviewResult = await codeReviewService.reviewCodeFromRepomix(repomixOutput, reviewOptions);
        
        // Format the response
        return {
          content: [
            { 
              type: 'text',
              text: JSON.stringify(reviewResult, null, 2) 
            }
          ]
        };
      } catch (error) {
        console.error('Error initializing code review service:', error);
        return {
          content: [
            { 
              type: 'text',
              text: `Error initializing code review service: ${(error as Error).message}. Make sure you have set the necessary environment variables (LLM_PROVIDER and the corresponding API key).`
            }
          ],
          isError: true
        };
      }
    } catch (error) {
      console.error('Error in code review:', error);
      return {
        content: [
          { 
            type: 'text',
            text: `Error performing code review: ${(error as Error).message}`
          }
        ],
        isError: true
      };
    }
  }

• src/index.ts:58-64 (schema)

  Zod schema defining the input parameters for the 'code_review' tool, including repoPath, optional specificFiles, fileTypes, detailLevel, and focusAreas.

  const codeReviewParams = {
    repoPath: z.string().describe('Path to the repository to analyze'),
    specificFiles: z.array(z.string()).optional().describe('Specific files to review'),
    fileTypes: z.array(z.string()).optional().describe('File types to include in the review'),
    detailLevel: z.enum(['basic', 'detailed']).optional().describe('Level of detail for the code review'),
    focusAreas: z.array(z.enum(['security', 'performance', 'quality', 'maintainability'])).optional().describe('Areas to focus on during the code review')
  };

• src/index.ts:67-129 (registration)

  Registration of the 'code_review' tool on the MCP server using server.tool(), including name, description, parameter schema, and handler function.

  server.tool(
    'code_review',
    'Use this tool when you need a comprehensive code review with specific feedback on code quality, security issues, performance problems, and maintainability concerns. This tool performs in-depth analysis on a repository or specific files and returns structured results including issues found, their severity, recommendations for fixes, and overall strengths of the codebase. Use it when you need actionable insights to improve code quality or when evaluating a codebase for potential problems.',
    codeReviewParams,
    async (params) => {
      try {
        // Execute Repomix to get the flattened codebase
        const repomixOptions: RepomixOptions = {
          includePaths: params.specificFiles || [params.repoPath],
          fileTypes: params.fileTypes,
          outputFormat: 'plain',
        };
        
        const repomixOutput = await executeRepomix(repomixOptions);
        
        // Set up review options
        const reviewOptions: CodeReviewOptions = {
          detailLevel: params.detailLevel || 'detailed',
          focusAreas: params.focusAreas || ['security', 'performance', 'quality', 'maintainability'],
        };
        
        // Create the code review service
        try {
          const codeReviewService = createCodeReviewService();
          
          // Perform the code review
          const reviewResult = await codeReviewService.reviewCodeFromRepomix(repomixOutput, reviewOptions);
          
          // Format the response
          return {
            content: [
              { 
                type: 'text',
                text: JSON.stringify(reviewResult, null, 2) 
              }
            ]
          };
        } catch (error) {
          console.error('Error initializing code review service:', error);
          return {
            content: [
              { 
                type: 'text',
                text: `Error initializing code review service: ${(error as Error).message}. Make sure you have set the necessary environment variables (LLM_PROVIDER and the corresponding API key).`
              }
            ],
            isError: true
          };
        }
      } catch (error) {
        console.error('Error in code review:', error);
        return {
          content: [
            { 
              type: 'text',
              text: `Error performing code review: ${(error as Error).message}`
            }
          ],
          isError: true
        };
      }
    }
  );

• src/llm/codeReviewService.ts:51-56 (helper)

  Core helper method in CodeReviewService that processes Repomix output and delegates to the private reviewCode method, which builds prompts and calls the LLM service for analysis.

  async reviewCodeFromRepomix(repomixOutput: string, options: CodeReviewOptions): Promise<CodeReviewResult> {
    console.log('Processing Repomix output...');
    const processedRepo = await this.codeProcessor.processRepomixOutput(repomixOutput);
    console.log(`Processed Repomix output (${processedRepo.length} characters)`);
    return this.reviewCode(processedRepo, options);
  }

• src/llm/codeReviewService.ts:64-83 (helper)

  Private helper method that constructs the code review prompt and invokes the LLM service with retry logic to generate the review result.

  private async reviewCode(code: string, options: CodeReviewOptions): Promise<CodeReviewResult> {
    try {
      console.log('Building code review prompt...');
      const prompt = this.promptBuilder.buildCodeReviewPrompt(code, options);
      
      console.log('Sending code to LLM for review...');
      // Use retry logic for robustness
      const result = await callWithRetry(
        () => this.llmService.generateReview(prompt),
        3,  // max retries
        2000 // initial delay
      );
      
      console.log('Review completed successfully');
      return result;
    } catch (error) {
      console.error('Error reviewing code:', error);
      throw new Error(`Failed to review code: ${(error as Error).message}`);
    }
  }