Deployment Cookbook — Documentation skill for Claude Code

Deployment Cookbook

Reference implementations for running LangChain agents in production. Each example is a full-stack chat app — streaming UI, subagents, thread history — deployed on a different platform using the same [Agent Streaming Protocol](https://github.com/langchain-ai/agent-protocol/tree/main/streaming).

Use these projects as starting points when you need to ship an agent-backed product: copy the stack that matches your hosting environment, swap in your own tools and models, and upgrade persistence when you move beyond a single instance.

What goes into an agent deployment

Every example in this cookbook follows the same shape. The framework and hosting change; the responsibilities do not.

1. Agent runtime

The agent itself, typically a LangGraph graph or [`deepagents`](https://www.npmjs.com/package/deepagents) coordinator, with tools, optional subagents, and middleware. It is compiled with a **checkpointer** so conversation state survives across turns. Examples start with an in-memory `MemorySaver` for simplicity; production deployments swap in Redis ([`@langchain/langgraph-checkpoint-redis`](https://www.npmjs.com/package/@langchain/langgraph-checkpoint-redis)), Postgres ([`@langchain/langgraph-checkpoint-postgres`](https://www.npmjs.com/package/@langchain/langgraph-checkpoint-postgres)), SQLite ([`@langchain/langgraph-checkpoint-sqlite`](https://www.npmjs.com/package/@langchain/langgraph-checkpoint-sqlite)), or platform-specific storage.

2. Protocol server

HTTP route handlers implement the [Agent Streaming Protocol](https://github.com/langchain-ai/agent-protocol/tree/main/streaming) under `/api/threads/...`.

Minimum (streaming chat)

These three endpoints are enough to run a single-threaded streaming chat with `HttpAgentServerAdapter`:

Method	Path	Purpose