Langgraph Toolkit is a TypeScript toolkit that helps you build AI agents, workflows, and stateful processes without starting with a bunch of configs, boilerplate, or infrastructure setup.

Start from zero config

Want to create a workflow? You just need to createWorkflow(), add nodes, compile(), and invoke().

No need to declare state if it is not required. No need to choose a model provider. No database, MCP server, checkpoint store, or web framework is needed to run the first workflow.

As your application grows, you can add only the capabilities you need.

From simple workflows to production AI systems

Langgraph Toolkit provides pre-built building blocks to extend workflows:

1. Typed workflows and states with nodes, edges, conditional routing, parallel, map, reduce, and subgraphs.
2. Streaming and observable execution for the frontend to monitor each step of processing, tool calls, reasoning, interruptions, and results in real-time.
3. Human in the loop with approval, interruptions, checkpoints, and resume() for workflows that require human review or need to continue processing later.
4. MCP native to connect tools, resources, prompts, database MCPs, and multiple MCP servers without stuffing connection logic into the graph.
5. Production-ready integrations with persistence through Redis, PostgreSQL, MySQL, SQLite, MongoDB, along with adapters for NestJS, Express, Fastify, and StruxJS.

One workflow, many environments

Workflows are written independently of the framework.

You can run the same resource in a worker, CLI, test, NestJS, Express, Fastify, or StruxJS without rewriting business logic.

Models are not locked into one vendor. You can use OpenAI compatible APIs, Hugging Face, Ollama, vLLM, TGI, LiteLLM, or your own provider.

Who should use it?

Langgraph Toolkit is suitable for TypeScript developers, backend engineers, and AI engineers building:

AI assistants, AI agents, multi-agent systems, RAG applications, database chats, automation workflows, background jobs, approval flows, or systems that use MCP.

It is especially suitable if you want the power of graph-based workflows but do not want the project to start with a load of abstractions and configurations.

How to get started

Install Core:

npm install @langgraph-toolkit/core

Create a workflow, add nodes, compile, and invoke.

Then, only install MCP, Community, Checkpointer, or framework adapters when the application really needs them.

The difference lies in the philosophy

Langgraph Toolkit does not try to turn every project into a complex AI architecture from day one.

Start with zero config. Add complexity only when you need it.

You can begin with a few lines of TypeScript, but still have a clear upgrade path to streaming, MCP, persistence, human approval, RAG, multi-agent, and workflow production without needing to rewrite the initial architecture.