LangChain said Nov. 19 it had opened LangSmith Tool Gateway in public beta, a managed service for controlling how AI agents access tools and credentials.LangChain’s announcement1

The service is designed to replace authentication and access-control plumbing built separately for each agent with a shared gateway. LangChain says teams can set which tools agents may call, choose whose credentials they use, and review records of calls.

The product arrives as companies develop agents that run on schedules or in response to events, rather than only answering prompts. Automatica reported Google’s Gemini API preview of tool retrieval3 on Nov. 19 as addressing a scaling bottleneck in complex agents; LangChain’s announcement focuses on governing tool access and identity, not tool retrieval.

LangChain described a large U.S. retailer building agents for store operations and inventory management. The company said the retailer’s engineering team had built more than 50 Model Context Protocol servers exposing over 1,000 tools. The example illustrates the problem LangChain is targeting, but the announcement does not say the retailer has deployed Tool Gateway.

The gateway lets teams register remote MCP servers or use built-in servers for services including Google, Microsoft, Salesforce, GitHub and Linear. Admins can bundle selected tools into a kit served through one MCP URL, so an agent sees the tools in that kit rather than every tool available to the workspace. The company says agents can also search a kit and load tools relevant to a task.

For each call, the gateway can use the individual user’s credentials or credentials associated with an agent or shared profile. LangChain’s documentation2 says the service identifies the caller, checks for a matching deny policy, attaches stored credentials or prompts the caller to connect, and seeks approval when a policy requires it before forwarding the call to an MCP server.

LangChain says calls can be traced with information such as the tool, server, user, outcome and duration; data sent to or returned by the tool is not stored in those records. The company says organization-wide audit logs are available on Enterprise plans, and its announcement directs customers to contact its team for access to deny policies and the audit log. It lists compatibility with MCP clients including LangChain, Claude Code and Cursor.

LangChain’s announcement gives no pricing, and it includes no independent test results.