// Import necessary modules from the AI SDK and OpenAI provider
import {
streamText,
createDataStreamResponse,
appendClientMessage,
type Message,
type DataStreamWriter,
type CoreUserMessage,
type CoreAssistantMessage,
type CoreSystemMessage
} from 'ai';
import { openai } from '@ai-sdk/openai';
import { createMcpClients } from '../../../lib/mcp-clients';
// Set the maximum duration for the response
export const maxDuration = 800;
// Define the POST handler
export async function POST(request: Request) {
try {
// Parse the incoming JSON request
const json = await request.json();
const { messages, message } = json;
console.log('Received request with message:', message);
// Initialize the OpenAI model
const model = openai('gpt-4o');
// Merge the new user message with existing messages
const fullMessages: Message[] = appendClientMessage({
messages: Array.isArray(messages) ? messages : [],
message,
});
console.log('Processing with full messages count:', fullMessages.length);
// Debug the message format
console.log('Message format check:', JSON.stringify(fullMessages));
// Create MCP clients for all enabled servers in the registry
console.log('Creating MCP clients from registry');
const { tools } = await createMcpClients();
console.log('Retrieved tools:', Object.keys(tools).length ? `${Object.keys(tools).length} tools available` : 'No tools available');
// Create and return the data stream response
return createDataStreamResponse({
execute: async (dataStream) => {
console.log('Starting to stream response');
try {
// Convert to proper CoreMessage format that streamText expects
const formattedMessages: (CoreUserMessage | CoreAssistantMessage | CoreSystemMessage)[] = [];
for (const msg of fullMessages) {
if (typeof msg === 'object' && msg !== null) {
if ('role' in msg && 'content' in msg) {
const role = msg.role;
const content = typeof msg.content === 'string' ? msg.content : String(msg.content);
if (role === 'user') {
formattedMessages.push({ role: 'user', content });
} else if (role === 'assistant') {
formattedMessages.push({ role: 'assistant', content });
} else if (role === 'system') {
formattedMessages.push({ role: 'system', content });
}
}
} else if (typeof msg === 'string') {
formattedMessages.push({ role: 'user', content: msg });
}
}
// Ensure we have at least one message
if (formattedMessages.length === 0) {
formattedMessages.push({ role: 'user', content: 'Hello' });
}
console.log('Formatted messages for LLM:', JSON.stringify(formattedMessages));
const result = await streamText({
model,
messages: formattedMessages,
tools,
toolChoice: 'auto', // Options: 'auto', 'required', 'manual'
maxSteps: 20,
});
console.log('Stream created successfully, merging into data stream');
// Merge the result into the data stream
await result.mergeIntoDataStream(dataStream, {
sendReasoning: true,
});
console.log('Stream merged successfully');
} catch (streamError: unknown) {
console.error('Error in streaming:', streamError);
const errorMessage = streamError instanceof Error ? streamError.message : 'Unknown error';
console.error('Streaming error details:', errorMessage);
// Return a simple error text directly without trying to use dataStream methods
// The SDK will automatically handle stream closing
}
},
});
} catch (error: any) {
console.error('Chat error:', error);
return new Response(error.message || 'Server error', { status: 500 });
}
}
