LangChain launches LangSmith Tool Gateway in public beta to govern agent tool calls
The managed layer routes agent tool requests through one MCP URL, handling credentials, identity, deny and approval policies, and tracing for each call.
What happened
LangChain has released Tool Gateway, a feature of its LangSmith platform. It sits between AI agents and the tools they use, such as Google, Salesforce or GitHub. It decides which tools an agent may call and whose login it uses.
Why it matters to you
Agents increasingly run on schedules without anyone watching. Companies need to control what they can do and keep a record of it.
The takeaway
Tool Gateway gives teams one place to control and audit what their AI agents are allowed to do.
What it means for you
You can add tool access once, publish a kit URL, and point any MCP client such as Claude Code or Cursor at it without writing OAuth flows per agent.
Security and compliance teams get a single record of which tool ran and under whose credentials, though deny policies and audit logs need a conversation with LangChain.
LangChain extends LangSmith from model-call governance (LLM Gateway) to tool-call governance, with audit logging tied to Enterprise plans.