Complete AI Training
Start for Free

Your job's AI kit

Get your AI kit

Tell us who you are and what you do. We show you your kit right away and email you the link: skills, prompts, AI agents, MCP servers and courses for your job.

500+ jobs ready, and we make a kit for any other job. No payment needed to look.

Share

AI news ·

LangSmith launches tool gateway to govern agent tool use with identity and access controls

LangChain's LangSmith Tool Gateway launched in public beta to govern agent tool calls, unifying credential and identity management as agents begin acting for thousands of users with distinct permissions.

LangChain launched LangSmith Tool Gateway into public beta on November 19, 2026, giving engineering teams a managed layer to control credentials, identity, and access for every tool an AI agent calls. The release addresses a growing governance problem as agents move from chat-based interactions to scheduled, event-driven work where they act on behalf of hundreds or thousands of users, each with distinct permissions.

Tool Gateway sits inside the LangSmith Agent Platform alongside the existing LLM Gateway, which already governs model calls with spend limits, rate limits, and data redaction. Together, the two gateways provide a unified view of agent actions and spending. "Tool Gateway is a managed tool layer that handles credentials, identity, and access for every agent tool call, making it simple to govern the actions your agents take," the company said.

Why agent tool use breaks at scale

When agents serve large user bases in production, managing tokens and recording access for audits becomes orders of magnitude harder than local development. The quick approach - hard-coding OAuth flows per provider, storing tokens beside the agent, and adding routing for each new tool - falls apart as usage, tool count, and user numbers grow.

LangChain pointed to a large US retailer building agents for store operations and inventory management. The engineering team built over 50 MCP servers exposing more than a thousand tools. One agent answers associate questions using individual credentials to access inventory systems, schedules, and email. Another runs at 3 a.m. daily to reconcile stock using its own service credentials. Without a central governance layer, nothing prevents an agent from using the wrong credential or calling an unapproved tool, and the security team has no way to audit what happened.

Access: One URL, scoped tool kits, and identity resolution

Tool Gateway lets callers route all tool requests through a single URL. Teams register MCP servers once or enable built-in servers for common apps like Google, Salesforce, or GitHub. From there, they build tool kits - curated sets of tools published at one URL - and scope each agent to only the tools it needs. When an agent has hundreds of tools, a search mode lets it query its kit and load only the tools relevant to the current task, keeping latency low.

For identity, the gateway resolves every call to one of two modes. Acting as the user applies the end user's own credentials through an identity provider like Okta or Entra ID. If the user hasn't connected yet, the call returns a sign-in link, waits for authentication, and continues automatically - no extra code required. Acting as the agent uses credentials tied to the agent itself, either through agent auth configured in LangSmith or through shared profiles that workspace admins connect once and reuse across autonomous agents.

Control and visibility across every tool call

Admins set policies at the org or workspace level. Deny policies block tool use for specific users, service keys, or combinations, with listed exceptions. Approval policies require a person to confirm sensitive tool calls - such as writing to a data warehouse - before the agent proceeds.

Every tool call is traced inside LangSmith with context on the agent conversation. Traces show whose credentials were used, which server handled the call, latency, and outcome. Sensitive data sent to or from the tool is not stored. For Enterprise plan customers, an audit log captures which tool ran and as whom across the organization, letting security teams answer questions like what an agent did at 3 a.m. and which credentials it used without pulling logs from multiple platforms.

Built-in tools skip server and OAuth setup

Tool Gateway ships with built-in servers that LangSmith hosts for Google, Microsoft, Salesforce, GitHub, and Linear. Teams connect an account or add an API key once and add the tools to any kit - no server to run, no OAuth app to register. Additional built-ins for tools like Exa, Tavily, Apollo, and Ashby require an API key or public access.

Why this matters for IT, operations, and product teams

For teams shipping agents that act on behalf of users or run autonomously, Tool Gateway removes the need to build and maintain auth plumbing for each new tool. Instead of hand-rolling OAuth flows and token storage per agent, engineers register a server once and compose scoped tool kits. Security and compliance teams get a single record of every tool call, with identity context, without stitching together traces from separate systems. The result is faster agent deployment and a smaller surface area for credential misuse - two concerns that grow in lockstep as agent fleets scale.

Share