managed-deep-agents

You can manage LangSmith deep agents via the REST API by orchestrating their creation, execution, and monitoring. This handles agent workflows end-to-end, covering agents, threads, runs, and MCP server registration with streaming results.

Handling human-in-the-loop interrupts for deep agents involves using per-tool interrupt configurations within durable threads. The orchestration layer manages these interrupts directly during production-like agent runs via LangSmith endpoints.

Yes, you can stream runs and register MCP servers with LangSmith agents. The orchestration process supports streaming results for runs and provides API-first control for registering MCP servers before deploying agent trees.

Before deploying agent trees in LangSmith, you need static API keys and familiarity with MCP server flows. You also need tools for agent lifecycle management to properly register servers and start threads in production-like environments.

The best way to manage an agent repository in LangSmith Context Hub is through API-first control. This allows you to deploy agents, start durable threads, stream runs, and handle interrupts within a unified orchestration workflow.

Using static API keys for deep agent orchestration requires secure key management and familiarity with MCP server flows. It is suited for production-like environments but requires dedicated tools for full agent lifecycle management.