LangChain Adds Secure Connections to Managed Deep Agents

LangChain has released Connections in Managed Deep Agents version 0.7.0 and higher. This new feature shifts credential management out of local environment files and project builds, housing them securely within the LangSmith workspace. Developers can now rotate or revoke API keys and OAuth tokens without modifying their codebase or triggering a redeployment.

The system distinguishes between agent-owned and user-owned credentials. Agent-owned secrets are shared across all callers for general tasks like web searches using Tavily. Conversely, user-owned credentials resolve to the specific individual interacting with the agent at run time. This means actions like opening a GitHub issue or filing a ticket are attributed directly to the user's personal handle rather than a generic service account or bot.

Connections simplifies the complex plumbing of OAuth integration by handling authorization round-trips, token storage, and refresh logic automatically. The release includes a built-in catalog featuring GitHub and 22 other services. Additionally, the system supports Model Context Protocol servers, such as Linear, which can register OAuth clients automatically using only a server URL. If a user has not yet authorized a connection, the agent automatically pauses the run to request a grant before proceeding.

For local development, agent-owned connections resolve from environment variables, while user-owned connections authenticate the signed-in developer. To implement a connection, developers use the command-line interface to create the credential, retrieve it in their tool code using a single get function, and redeploy. This workflow eliminates the need to build custom callback routes or consent screens.