LangSmith Tool Gateway: One URL to Govern Every Agent Tool Call

LangSmith Tool Gateway: One URL to Govern Every Agent Tool Call

Agentic AI

One LangChain customer, a large US retailer, has built more than 50 MCP servers exposing over a thousand tools, and when credentials are hard-coded, nothing deterministic stops an agent from calling an unapproved tool or the wrong credential. LangSmith Tool Gateway, now in public beta, is LangChain’s answer for platform and security teams: put every agent tool call behind one managed URL that decides what the agent can reach, whose identity it uses, and what gets logged.

What Happened

LangChain announced Tool Gateway on the LangSmith Agent Platform as a managed tool layer that handles credentials, identity and access for agent tool calls. It sits next to LLM Gateway, which already governs model calls with spend limits, fallbacks, rate limits and redaction. Tool Gateway does the same job for actions.

The building block is a tool kit: a curated set of tools published at a single MCP URL, shared with a workspace or kept personal. An agent connected to a kit sees only the tools in it. Teams can register any MCP server by URL or switch on built-in servers that LangSmith hosts for Google, Microsoft, Salesforce, GitHub and Linear, with no server to run and no OAuth app to register. For kits with hundreds or thousands of tools, a search mode lets the agent query the kit and load only what the current task needs.

Why It Matters

LangChain frames the problem in three requirements: access (which tools and credentials), control (what each user may do, with review where needed), and visibility (proof the agent used the right credentials). The common shortcut is hard-coding all three per agent, with an OAuth flow per provider and tokens stored beside the agent. The post argues that this breaks once users, tools and agents multiply.

The retailer example shows why identity is the hard part. A store assistant answers associates’ questions using each associate’s own credentials. An inventory agent reconciles stock at 3 a.m. on service credentials. Tool Gateway maps every call to one of two identities. Acting as the user applies that person’s credentials through an identity provider such as Okta or Entra ID, and if they haven’t connected yet, the call returns a sign-in link and resumes after they authenticate. Acting as the agent uses agent auth configured in LangSmith or a profile with shared credentials that an admin connects once.

What This Means for Builders

The control layer is policy, not code. Deny policies block tools for a user, a service key or a combination, at org or workspace level, and can apply to everyone with listed exceptions. Approval policies make the user confirm a sensitive call in their MCP client before it runs, such as a write to a data warehouse.

Visibility lands in the trace you already debug with. Each call can record its tool, server, user, outcome and duration, while the data sent to or returned by the tool isn’t stored. On Enterprise plans, an audit log captures which tool ran and as whom across the organization. LangChain lists five setup steps, with details in the Tool Gateway docs: open Tools in LangSmith, add a server, connect accounts, build a kit, and point any MCP client at the kit URL. LangChain names LangChain, Deep Agents, Claude Code and Cursor as supported clients.

Two caveats are worth noting. It’s a beta, and the post says deny policies and the audit log require talking to LangChain’s team for access. It also extends the platform bet LangChain made with its agent registry, so teams not already on LangSmith would be adopting a platform, not just a proxy.

The So What

Platform teams running more than a handful of production agents on shared MCP servers should map which agents act as users and which as themselves, then pilot Tool Gateway with one kit per agent and an approval policy on every write tool. If deny policies and the audit log are the reason to adopt it, confirm access with LangChain before building around them.

Source: LangChain Blog

Content created with AI assistance and reviewed for accuracy.

💬

Join the conversation

Stack Insiders is our free community for readers who want to go deeper — share resources, ask questions, and connect with others across every vertical we cover.

Join Stack Insiders →

Newsletter coming soon.

Curated digests across AI, biohacking, photography, travel, and more. Be the first to know when we launch.