Ultra MCP

Instructions

Implementation Reference

• src/handlers/ai-tools.ts:222-263 (handler)

  Core handler function that executes the 'research' tool. Selects AI provider, constructs research-specific system prompt based on output format and sources, generates response via provider.generateText, and returns structured content with metadata.

  async handleResearch(params: z.infer<typeof ResearchSchema>) {
    // Use provided provider or get the preferred one (Azure if configured)
    const providerName = params.provider || (await this.providerManager.getPreferredProvider(['openai', 'gemini', 'azure', 'grok']));
    const provider = await this.providerManager.getProvider(providerName);
    
    // Build research prompts based on output format
    const formatInstructions = {
      summary: "Provide a concise summary of key findings and insights.",
      detailed: "Provide a comprehensive analysis with detailed findings, evidence, and conclusions.",
      academic: "Present findings in an academic format with clear structure, citations where possible, and scholarly analysis.",
    };
  
    const systemPrompt = `You are an expert researcher with deep knowledge across multiple domains. 
    Your task is to conduct thorough research, analyze information critically, and present findings clearly.
    ${params.sources ? `Consider these specific sources or contexts: ${params.sources.join(", ")}` : ""}
    ${formatInstructions[params.outputFormat]}`;
  
    const response = await provider.generateText({
      prompt: `Research the following: ${params.query}`,
      model: params.model,
      systemPrompt,
      reasoningEffort: (providerName === "openai" || providerName === "azure" || providerName === "grok") ? "high" : undefined,
      useSearchGrounding: providerName === "gemini", // Always enable search for research with Gemini
      temperature: 0.4, // Lower temperature for research accuracy
    });
  
    return {
      content: [
        {
          type: "text",
          text: response.text,
        },
      ],
      metadata: {
        provider: providerName,
        model: response.model,
        outputFormat: params.outputFormat,
        usage: response.usage,
        ...response.metadata,
      },
    };
  }

• src/server.ts:264-272 (registration)

  Registers the 'research' tool with MCP server, defining title, description, input schema, and handler delegation to AIToolHandlers.handleResearch.

  // Register research tool
  server.registerTool("research", {
    title: "Research",
    description: "Conduct comprehensive research with multiple output formats",
    inputSchema: ResearchSchema.shape,
  }, async (args) => {
    const aiHandlers = await getHandlers();
    return await aiHandlers.handleResearch(args);
  });

• src/server.ts:26-32 (schema)

  Zod schema defining input parameters for the 'research' tool: provider, query (required), sources, model, outputFormat.

  const ResearchSchema = z.object({
    provider: z.enum(["openai", "gemini", "azure", "grok"]).optional().describe("AI provider to use (defaults to Azure if configured, otherwise best available)"),
    query: z.string().describe("Research query or topic"),
    sources: z.array(z.string()).optional().describe("Specific sources or contexts to consider"),
    model: z.string().optional().describe("Specific model to use"),
    outputFormat: z.enum(["summary", "detailed", "academic"]).default("detailed").describe("Output format for research"),
  });

• src/handlers/ai-tools.ts:26-32 (schema)

  Zod schema for 'research' tool inputs, used in handler type inference (identical to server.ts definition).

  const ResearchSchema = z.object({
    provider: z.enum(["openai", "gemini", "azure", "grok"]).optional().describe("AI provider to use (defaults to Azure if configured, otherwise best available)"),
    query: z.string().describe("Research query or topic"),
    sources: z.array(z.string()).optional().describe("Specific sources or contexts to consider"),
    model: z.string().optional().describe("Specific model to use"),
    outputFormat: z.enum(["summary", "detailed", "academic"]).default("detailed").describe("Output format for research"),
  });

• src/server.ts:551-569 (registration)

  Registers a prompt template for natural language invocation of the 'research' tool, mapping args to a user message prompt.

  server.registerPrompt("research", {
    title: "Comprehensive Research",
    description: "Conduct thorough research on any topic with multiple output formats",
    argsSchema: {
      query: z.string().optional(),
      provider: z.string().optional(),
      model: z.string().optional(),
      outputFormat: z.string().optional(),
      sources: z.string().optional(),
    },
  }, (args) => ({
    messages: [{
      role: "user",
      content: {
        type: "text",
        text: `Research this topic thoroughly: ${args.query || 'Please specify a research topic or query.'}${args.outputFormat ? ` (format: ${args.outputFormat})` : ''}${args.sources ? `\n\nFocus on these sources: ${args.sources}` : ''}${args.provider ? ` (using ${args.provider} provider)` : ''}`
      }
    }]
  }));