vercel-ai-sdk

Streaming chat interfaces with Vercel AI SDK use useChat on the client and streamText on the server. useChat manages message state and handles streaming from your API route, which uses streamText with a model string and system prompt, then returns the streamed response via toTextStreamResponse for proper formatting.

Tool calling in v5 uses the tool() function with inputSchema and a handler function. Define tools as an array, pass them to streamText or generateText, and import tool definitions directly from the ai package. Zod schemas define inputSchema structure, and the SDK automatically handles tool invocation and result streaming.

Yes, Vercel AI SDK supports embeddings for semantic search and retrieval-augmented generation. Generate embeddings for your documents, store vectors, then retrieve relevant context during chat or text generation to augment model responses with external knowledge.

MCP integration connects your Vercel AI SDK application to MCP servers, enabling dynamic tool discovery and execution. Configure the MCP connection, and tools from the server become available within chat and text generation workflows for expanded capability without hardcoding tools.

Yes, Vercel AI SDK v5 supports multiple providers through string-based model specification. Pass provider-prefixed model strings like 'openai:gpt-4' or 'anthropic:claude-3' to streamText and generateText, allowing provider switching without code changes across chat, embeddings, and tool calling.

toUIMessageStreamResponse and toTextStreamResponse format streaming data correctly for client consumption. toUIMessageStreamResponse works with useChat by streaming structured messages; toTextStreamResponse handles plain text streaming. Both ensure proper client-side stream handling and error propagation.