Context Optimizer MCP Server

Instructions

Implementation Reference

• src/tools/askFollowUp.ts:21-51 (handler)

  The main execute handler that validates the input question, loads the previous terminal session, processes the follow-up using LLM, and returns success or error response.

  async execute(args: any): Promise<MCPToolResponse> {
    try {
      this.logOperation('Follow-up question started', { question: args.question });
      
      // Validate required fields
      const fieldError = this.validateRequiredFields(args, ['question']);
      if (fieldError) {
        return this.createErrorResponse(fieldError);
      }
      
      // Load previous terminal session
      const session = await SessionManager.loadTerminalSession();
      if (!session) {
        return this.createErrorResponse(
          'No recent terminal execution found. Please use the runAndExtract tool first to execute a command, then you can ask follow-up questions about its output.'
        );
      }
      
      // Process follow-up question with LLM
      const answer = await this.processFollowUpQuestion(session, args.question);
      
      this.logOperation('Follow-up question completed successfully');
      return this.createSuccessResponse(answer);
      
    } catch (error) {
      this.logOperation('Follow-up question failed', { error });
      return this.createErrorResponse(
        `Follow-up question failed: ${error instanceof Error ? error.message : String(error)}`
      );
    }
  }

• src/tools/askFollowUp.ts:10-19 (schema)

  Input schema defining the required 'question' parameter as a string.

  readonly inputSchema = {
    type: 'object',
    properties: {
      question: {
        type: 'string',
        description: 'Follow-up question about the previous terminal command execution and its output'
      }
    },
    required: ['question']
  };

• src/server.ts:60-74 (registration)

  Registers the AskFollowUpTool by instantiating it (line 64) and adding to the tools map used by the MCP server handlers.

  private setupTools(): void {
    const toolInstances = [
      new AskAboutFileTool(),
      new RunAndExtractTool(),
      new AskFollowUpTool(),
      new ResearchTopicTool(),
      new DeepResearchTool()
    ];
    
    for (const tool of toolInstances) {
      this.tools.set(tool.name, tool);
    }
    
    Logger.info(`Registered ${this.tools.size} tools: ${Array.from(this.tools.keys()).join(', ')}`);
  }

• src/tools/askFollowUp.ts:53-69 (helper)

  Helper method that creates the LLM prompt and calls the provider to get the answer based on session and question.

  private async processFollowUpQuestion(
    session: any,
    question: string
  ): Promise<string> {
    const config = ConfigurationManager.getConfig();
    const provider = LLMProviderFactory.createProvider(config.llm.provider);
    const apiKey = this.getApiKey(config.llm.provider, config.llm);
    
    const prompt = this.createFollowUpPrompt(session, question);
    const response = await provider.processRequest(prompt, config.llm.model, apiKey);
    
    if (!response.success) {
      throw new Error(`LLM processing failed: ${response.error}`);
    }
    
    return response.content;
  }

• src/tools/askFollowUp.ts:71-92 (helper)

  Helper that constructs the detailed prompt for the LLM including previous command details, output, and instructions.

    private createFollowUpPrompt(session: any, question: string): string {
      return `You are answering a follow-up question about a previous terminal command execution.
  
  Previous command: ${session.command}
  Working directory: ${session.workingDirectory}
  Exit code: ${session.exitCode}
  Original extraction request: ${session.extractionPrompt}
  Previously extracted information: ${session.extractedInfo}
  
  Follow-up question: ${question}
  
  Context - Original command output:
  ${session.output}
  
  Instructions:
  - Answer the follow-up question based on the command output and context
  - Reference the original output when needed
  - Be specific and helpful
  - If the question cannot be answered from the available information, say so clearly
  - Use markdown formatting for better readability
  - Be concise but thorough in your response`;
    }