MCP Gemini Server

import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js"; import { z } from "zod"; import { GeminiService } from "../services/index.js"; import { logger } from "../utils/index.js"; import { mapAnyErrorToMcpError } from "../utils/errors.js"; // Tool Name and Description export const GEMINI_URL_ANALYSIS_TOOL_NAME = "gemini_url_analysis"; export const GEMINI_URL_ANALYSIS_TOOL_DESCRIPTION = ` Advanced URL analysis tool that fetches content from web pages and performs specialized analysis tasks. Supports various analysis types including summarization, comparison, information extraction, and Q&A. Automatically handles URL fetching, content processing, and intelligent model selection for optimal results. `; // Analysis types enum const analysisTypeSchema = z .enum([ "summary", "comparison", "extraction", "qa", "sentiment", "fact-check", "content-classification", "readability", "seo-analysis", ]) .describe("Type of analysis to perform on the URL content"); // Extraction schema for structured data extraction const extractionSchemaSchema = z .record(z.unknown()) .optional() .describe( "JSON schema or structure definition for extracting specific information from content" ); // Parameters for the URL analysis tool export const GEMINI_URL_ANALYSIS_PARAMS = { urls: z .array(z.string().url()) .min(1) .max(20) .describe("URLs to analyze (1-20 URLs supported)"), analysisType: analysisTypeSchema, query: z .string() .min(1) .optional() .describe("Specific query or instruction for the analysis"), extractionSchema: extractionSchemaSchema, questions: z .array(z.string()) .optional() .describe("List of specific questions to answer (for Q&A analysis)"), compareBy: z .array(z.string()) .optional() .describe("Specific aspects to compare when using comparison analysis"), outputFormat: z .enum(["text", "json", "markdown", "structured"]) .default("text") .optional() .describe("Desired output format for the analysis results"), includeMetadata: z .boolean() .default(true) .optional() .describe( "Include URL metadata (title, description, etc.) in the analysis" ), fetchOptions: z .object({ maxContentKb: z .number() .min(1) .max(1000) .default(100) .optional() .describe("Maximum content size per URL in KB"), timeoutMs: z .number() .min(1000) .max(30000) .default(10000) .optional() .describe("Fetch timeout per URL in milliseconds"), allowedDomains: z .array(z.string()) .optional() .describe("Specific domains to allow for this request"), userAgent: z .string() .optional() .describe("Custom User-Agent header for URL requests"), }) .optional() .describe("Advanced options for URL fetching"), modelName: z .string() .optional() .describe("Specific Gemini model to use (auto-selected if not specified)"), }; /** * Registers the gemini_url_analysis tool with the MCP server. * Provides specialized URL analysis capabilities with intelligent content processing. */ export const geminiUrlAnalysisTool = ( server: McpServer, serviceInstance: GeminiService ): void => { const processRequest = async (args: unknown) => { // Parse and validate the arguments const parsedArgs = z.object(GEMINI_URL_ANALYSIS_PARAMS).parse(args); logger.debug(`Received ${GEMINI_URL_ANALYSIS_TOOL_NAME} request:`, { urls: parsedArgs.urls, analysisType: parsedArgs.analysisType, urlCount: parsedArgs.urls.length, }); try { const { urls, analysisType, query, extractionSchema, questions, compareBy, outputFormat, includeMetadata, fetchOptions, modelName, } = parsedArgs; // Build the analysis prompt based on the analysis type const prompt = buildAnalysisPrompt({ analysisType, query, extractionSchema, questions, compareBy, outputFormat, urlCount: urls.length, }); // Prepare URL context for content generation const urlContext = { urls, fetchOptions: { ...fetchOptions, includeMetadata: includeMetadata ?? true, convertToMarkdown: true, // Always convert to markdown for better analysis }, }; // Calculate URL context metrics for optimal model selection const urlCount = urls.length; const maxContentKb = fetchOptions?.maxContentKb || 100; const estimatedUrlContentSize = urlCount * maxContentKb * 1024; // Select task type based on analysis type const taskType = getTaskTypeForAnalysis(analysisType); // Generate analysis using the service const analysisResult = await serviceInstance.generateContent({ prompt, modelName, urlContext, taskType: taskType as | "text-generation" | "image-generation" | "video-generation" | "code-review" | "multimodal" | "reasoning", preferQuality: true, // Prefer quality for analysis tasks complexityHint: urlCount > 5 ? "complex" : "medium", urlCount, estimatedUrlContentSize, systemInstruction: getSystemInstructionForAnalysis( analysisType, outputFormat ), }); // Format the result based on output format const formattedResult = formatAnalysisResult( analysisResult, outputFormat ); return { content: [ { type: "text" as const, text: formattedResult, }, ], }; } catch (error: unknown) { logger.error(`Error processing ${GEMINI_URL_ANALYSIS_TOOL_NAME}:`, error); throw mapAnyErrorToMcpError(error, GEMINI_URL_ANALYSIS_TOOL_NAME); } }; // Register the tool with the server server.tool( GEMINI_URL_ANALYSIS_TOOL_NAME, GEMINI_URL_ANALYSIS_TOOL_DESCRIPTION, GEMINI_URL_ANALYSIS_PARAMS, processRequest ); logger.info(`Tool registered: ${GEMINI_URL_ANALYSIS_TOOL_NAME}`); }; /** * Builds the analysis prompt based on the requested analysis type and parameters */ function buildAnalysisPrompt(params: { analysisType: string; query?: string; extractionSchema?: Record<string, unknown>; questions?: string[]; compareBy?: string[]; outputFormat?: string; urlCount: number; }): string { const { analysisType, query, extractionSchema, questions, compareBy, outputFormat, urlCount, } = params; let prompt = `Perform a ${analysisType} analysis on the provided URL content${urlCount > 1 ? "s" : ""}.\n\n`; switch (analysisType) { case "summary": prompt += `Provide a comprehensive summary of the main points, key information, and important insights from the content. `; if (query) { prompt += `Focus particularly on: ${query}. `; } break; case "comparison": if (urlCount < 2) { prompt += `Since only one URL is provided, analyze the different aspects or sections within the content. `; } else { prompt += `Compare and contrast the content from the different URLs, highlighting similarities, differences, and unique aspects. `; } if (compareBy && compareBy.length > 0) { prompt += `Focus your comparison on these specific aspects: ${compareBy.join(", ")}. `; } break; case "extraction": prompt += `Extract specific information from the content. `; if (extractionSchema) { prompt += `Structure the extracted information according to this schema: ${JSON.stringify(extractionSchema, null, 2)}. `; } if (query) { prompt += `Focus on extracting: ${query}. `; } break; case "qa": prompt += `Answer the following questions based on the content:\n`; if (questions && questions.length > 0) { questions.forEach((question, index) => { prompt += `${index + 1}. ${question}\n`; }); } else if (query) { prompt += `Question: ${query}\n`; } else { prompt += `Provide comprehensive answers to common questions that would arise from this content.\n`; } break; case "sentiment": prompt += `Analyze the sentiment and emotional tone of the content. Identify the overall sentiment (positive, negative, neutral) and specific emotional indicators. `; if (query) { prompt += `Pay special attention to sentiment regarding: ${query}. `; } break; case "fact-check": prompt += `Evaluate the factual accuracy and credibility of claims made in the content. Identify verifiable facts, questionable claims, and potential misinformation. `; if (query) { prompt += `Focus particularly on claims about: ${query}. `; } break; case "content-classification": prompt += `Classify and categorize the content by topic, type, audience, and other relevant dimensions. `; if (query) { prompt += `Use this classification framework: ${query}. `; } break; case "readability": prompt += `Analyze the readability, writing quality, and accessibility of the content. Evaluate complexity, clarity, structure, and target audience. `; break; case "seo-analysis": prompt += `Perform an SEO analysis of the content, evaluating keyword usage, content structure, meta information, and optimization opportunities. `; break; default: if (query) { prompt += `Based on the following instruction: ${query}. `; } } // Add output format instructions if (outputFormat && outputFormat !== "text") { switch (outputFormat) { case "json": prompt += `\n\nFormat your response as valid JSON with appropriate structure and fields.`; break; case "markdown": prompt += `\n\nFormat your response in well-structured Markdown with appropriate headers, lists, and formatting.`; break; case "structured": prompt += `\n\nOrganize your response in a clear, structured format with distinct sections and subsections.`; break; } } prompt += `\n\nBe thorough, accurate, and insightful in your analysis.`; return prompt; } /** * Maps analysis types to task types for model selection */ function getTaskTypeForAnalysis(analysisType: string): string { switch (analysisType) { case "comparison": case "fact-check": case "seo-analysis": return "reasoning"; case "extraction": case "content-classification": return "text-generation"; default: return "text-generation"; } } /** * Generates system instructions based on analysis type and output format */ function getSystemInstructionForAnalysis( analysisType: string, outputFormat?: string ): string { let instruction = `You are an expert content analyst specializing in ${analysisType} analysis. `; switch (analysisType) { case "summary": instruction += `Provide concise yet comprehensive summaries that capture the essence and key insights of the content.`; break; case "comparison": instruction += `Excel at identifying similarities, differences, and patterns across different content sources.`; break; case "extraction": instruction += `Focus on accurately identifying and extracting specific information while maintaining context and relevance.`; break; case "qa": instruction += `Provide clear, accurate, and well-supported answers based on the available content.`; break; case "sentiment": instruction += `Accurately identify emotional tone, sentiment indicators, and subjective language patterns.`; break; case "fact-check": instruction += `Evaluate claims critically, distinguish between facts and opinions, and identify potential misinformation.`; break; case "content-classification": instruction += `Categorize content accurately using relevant taxonomies and classification frameworks.`; break; case "readability": instruction += `Assess content accessibility, complexity, and effectiveness for target audiences.`; break; case "seo-analysis": instruction += `Evaluate content from an SEO perspective, focusing on optimization opportunities and best practices.`; break; } if (outputFormat === "json") { instruction += ` Always respond with valid, well-structured JSON.`; } else if (outputFormat === "markdown") { instruction += ` Use proper Markdown formatting with clear headers and structure.`; } instruction += ` Base your analysis strictly on the provided content and clearly distinguish between what is explicitly stated and what is inferred.`; return instruction; } /** * Formats the analysis result based on the requested output format */ function formatAnalysisResult(result: string, outputFormat?: string): string { if (!outputFormat || outputFormat === "text") { return result; } // For other formats, the formatting should have been handled by the model // based on the prompt instructions, so we return the result as-is return result; }