Perplexity MCP Server

Instructions

Implementation Reference

• src/handlers/SearchHandler.ts:22-133 (handler)

  The handle method implements the 'search' tool execution: validates input, selects appropriate Perplexity model based on query intent, applies domain/recency filters, performs API search (streaming or standard), and returns formatted response with model info and cited sources.
  async handle(request: McpRequest) { if (!isValidSearchArgs(request.params.arguments)) { throw ErrorHandler.createMcpError( ErrorCode.InvalidParams, "Invalid search arguments. Query must be a non-empty string." ); } const { query, stream = false } = request.params.arguments as SearchArgs; try { // Get model selection based on intent const selection = this.modelSelectionService.selectModelBasedOnIntent(query); let description: string; let selectedModel: string; // Determine which model to use if (this.useAutoSelection) { // Auto-select model based on intent selectedModel = selection.model; description = selection.description; } else { // We're using a manually set model // But check if we have a strong match that should override if (selection.score >= 2 && selection.model !== this.currentModel) { // Strong intent match - override manual selection selectedModel = selection.model; description = `${selection.description} (auto-selected based on query intent)`; } else { // Stick with manually selected model selectedModel = this.currentModel; description = this.modelSelectionService.getModelDescription(this.currentModel); } } // Get domain filters and recency filter const domainFilterArray = this.filterState.getDomainFilterArray(); const recencyFilter = this.filterState.getRecencyFilter(); // Create API parameters const apiParams = this.apiService.createApiParams( selectedModel, query, domainFilterArray.length > 0 ? domainFilterArray : undefined, recencyFilter || undefined, stream ); // Include model information at the start of responses const modelInfo = `[Using model: ${selectedModel} - ${description}]\n\n`; if (stream) { // Handle streaming response const streamResult = await this.apiService.searchStream(apiParams); let responseText = modelInfo + streamResult.content; // Add citations if available if (streamResult.search_results && streamResult.search_results.length > 0) { responseText += '\n\n## Sources\n'; streamResult.search_results.forEach((result, index) => { const dateInfo = result.date ? ` (${result.date})` : ''; responseText += `[${index + 1}] ${result.title}${dateInfo}\n${result.url}\n\n`; }); } return { content: [ { type: "text", text: responseText, }, ], }; } else { // Handle regular response const response = await this.apiService.search(apiParams); if (response.choices && response.choices.length > 0) { let responseText = modelInfo + response.choices[0].message.content; // Add citations if available if (response.search_results && response.search_results.length > 0) { responseText += '\n\n## Sources\n'; response.search_results.forEach((result, index) => { const dateInfo = result.date ? ` (${result.date})` : ''; responseText += `[${index + 1}] ${result.title}${dateInfo}\n${result.url}\n\n`; }); } return { content: [ { type: "text", text: responseText, }, ], }; } throw new Error("No response content received"); } } catch (error) { if (error instanceof Error && error.message === "No response content received") { throw error; } // Handle API errors return stream ? ErrorHandler.handleStreamingApiError(error) : ErrorHandler.handleApiError(error); } }
• src/schemas/toolSchemas.ts:8-25 (schema)

  Input schema definition for the 'search' tool, specifying query as required string and optional stream boolean.
  { name: "search", description: "Web search via Perplexity AI with automatic model selection. Returns cited sources with summaries. The search uses only the query text (not conversation history). Best for: current events, factual research, technical documentation, comparative analysis.", inputSchema: { type: "object", properties: { query: { type: "string", description: "Direct search query. Be specific with 2-3 context words, use expert terminology. Good: 'Compare 2025 React vs Vue performance for enterprise apps'. Bad: 'tell me about frameworks'. Tips: Use 'site:domain.com' for specific sites, include years for recent info, add 'analyze/compare/explain' for reasoning tasks.", }, stream: { type: "boolean", description: "Enable streaming responses (default: false)", }, }, required: ["query"], }, },
• src/server/PerplexityMcpServer.ts:106-136 (registration)

  Registration of tool handlers in the MCP server's call tool request handler. The 'search' tool is dispatched to SearchHandler.handle() based on request.params.name.
  this.server.setRequestHandler(CallToolRequestSchema, async (request) => { // Update search handler with current model state this.searchHandler = new SearchHandler( this.apiService, this.modelSelectionService, this.filterState, this.currentModel, this.useAutoSelection ); switch (request.params.name) { case "search": return this.searchHandler.handle(request); case "domain_filter": return this.domainFilterHandler.handle(request); case "recency_filter": return this.recencyFilterHandler.handle(request); case "clear_filters": return this.filterManagementHandler.handleClearFilters(); case "list_filters": return this.filterManagementHandler.handleListFilters(); case "model_info": return this.modelInfoHandler.handle(request); default: throw new McpError( ErrorCode.MethodNotFound, `Unknown tool: ${request.params.name}` ); } }); }
• src/server/PerplexityMcpServer.ts:101-103 (registration)

  Registration of tool schemas via ListToolsRequestSchema handler, returning TOOL_SCHEMAS array which includes the 'search' tool schema.
  this.server.setRequestHandler(ListToolsRequestSchema, async () => ({ tools: TOOL_SCHEMAS, }));