Raindrop.io

// External imports import { OpenAI } from 'openai'; import { z } from 'zod'; import type { MCPResponse } from '../types/mcp'; // Internal imports /** * Zod schema for OpenAI ChatGPT request */ export const chatRequestSchema = z.object({ model: z.string().default('gpt-3.5-turbo'), messages: z.array(z.object({ role: z.enum(['system', 'user', 'assistant']), content: z.string(), })).min(1, { message: "At least one message is required" }), temperature: z.number().min(0).max(2).optional(), max_tokens: z.number().min(1).max(4096).optional(), }); /** * Zod schema for OpenAI ChatGPT response */ export const chatResponseSchema = z.object({ id: z.string(), object: z.string(), created: z.number(), choices: z.array(z.object({ index: z.number(), message: z.object({ role: z.string(), content: z.string(), }), finish_reason: z.string().optional(), })), usage: z.object({ prompt_tokens: z.number(), completion_tokens: z.number(), total_tokens: z.number(), }).optional(), }); export interface ChatRequest extends z.infer<typeof chatRequestSchema> { } export interface ChatResponse extends z.infer<typeof chatResponseSchema> { } // Remove module-level cache function getOpenAIClient(apiKey: string): OpenAI { // Always instantiate a new client to respect test mocks return new OpenAI({ apiKey }); } // Defensive MCPError factory for protocol-compliant error responses export function MCPError(message: string): { success: false; error: string } { return { success: false, error: message }; } /** * Sends a chat message to OpenAI's ChatGPT API and returns the response. * @param request ChatRequest object * @param apiKey OpenAI API key * @returns ChatResponse object */ export async function sendChatMessage( request: ChatRequest, apiKey: string ): Promise<MCPResponse<ChatResponse> | { success: false; error: string }> { try { const validatedRequest = chatRequestSchema.parse(request); const openai = getOpenAIClient(apiKey); // Defensive: Ensure the client structure matches expectations if ( !openai || !('chat' in openai) || !openai.chat || !('completions' in openai.chat) || !openai.chat.completions || typeof openai.chat.completions.create !== 'function' ) { throw new Error( 'OpenAI client is not properly initialized or does not match expected structure. ' + 'Check your OpenAI library version and test mocks.' ); } // Only include max_tokens if defined, and ensure it's a number const completionParams = { model: validatedRequest.model, messages: validatedRequest.messages, temperature: validatedRequest.temperature ?? null, ...(typeof validatedRequest.max_tokens === 'number' ? { max_tokens: validatedRequest.max_tokens } : {}) }; const response = await openai.chat.completions.create(completionParams); const validatedResponse = chatResponseSchema.parse(response); return { success: true, data: validatedResponse }; } catch (error: any) { // Always return protocol-compliant error response return MCPError(`OpenAI API error: ${error?.message || error}`); } } /** * Documentation: * - Set your OpenAI API key in your environment and pass it to sendChatMessage. * - Uses official OpenAI library for API calls. * - Use MCPResponse and MCPError for protocol-compliant responses. * - See MCP protocol docs for stdio transport and tool integration. */