Build LangChain TypeScript App

Build LangChain.js v1 and LangGraph.js applications in TypeScript. Choose one implementation path before coding, keep the spine small, and load only the bundled references the chosen path needs.

When to use this skill

• Building a tool-calling assistant with createAgent from langchain and Zod-typed tools.
• Wiring a RAG pipeline with a Document loader, splitter, embeddings, vector store, and retriever.
• Authoring a raw StateGraph from @langchain/langgraph with explicit state, routing, fan-out/fan-in, or interrupts.
• Adding structured output via withStructuredOutput, responseFormat, toolStrategy, or providerStrategy.
• Streaming tokens, events, or graph updates back to a UI/API and handling cancellation.
• Persisting agent/graph state with checkpointers, stores, or thread_id-keyed memory.
• Connecting MCP servers as namespaced tools through @langchain/mcp-adapters.
• Composing supervisor/router/handoff multi-agent or knowledge-domain agents in TypeScript.

Do not use this skill when:

• The codebase is Python LangChain / LangGraph — TypeScript-only patterns here will mislead.
• A simple single-provider chat call (e.g., openai.chat.completions.create) is enough — pull in the provider SDK directly.