LangChain launches LangSmith Tool Gateway in public beta
LangChain launched LangSmith Tool Gateway in public beta to mediate AI agent tool calls with credentials, access policies, caller approvals and execution records.
LangChain launched LangSmith Tool Gateway in public beta on November 19, adding a managed control layer between AI agents and tools exposed through Model Context Protocol servers. The LangSmith Agent Platform feature centralizes credentials, caller identities, access policies, approvals and tool-call records.
LangChain describes Tool Gateway as a managed layer that routes tool requests through one URL. For each caller, the service resolves identity, manages access and logs tool-call traces. These are documented product mechanics; LangChain’s claims that the gateway makes controlled access easier and faster have not been independently tested.
The gateway can apply a user’s credentials when an agent acts on that person’s behalf. For unattended work, it can instead use credentials tied to an agent or to an administrator-configured profile. LangChain also says it offers hosted built-in servers for services including Google, Microsoft, Salesforce, GitHub and Linear.
Administrators can publish a curated set of tools at one MCP address through a tool kit, limiting the tools presented to a connected agent. LangChain says administrators can also create deny policies for users or service keys at the organization or workspace level. Approval policies prompt a user in the MCP client to confirm a sensitive call before it runs.
Those execution-time controls are one layer of governance; metadata-based agent governance addresses the separate task of classifying agents and MCP servers.
When tool-call tracing is enabled, LangChain says the record includes the tool, server, user, outcome and duration. The company says data sent to or returned by the tool is not stored in that record. On Enterprise plans, an audit log records which tool ran and whose identity it used.
The announcement does not provide an independent security evaluation, an identified customer deployment or a general-availability date. Its claims about easier administration, faster setup and improved security visibility remain vendor assertions rather than independently demonstrated outcomes.
More news

Perplexity launches Monitor API for recurring web alerts

Perplexity launches Context API for multi-page web extraction

Google launches Gemini 4 Neon across API, business and paid consumer tiers
