managed-deep-agents

What problem does it solve?

This Skill eliminates the heavy operational work of standing up and operating Deep Agents by providing an API-first managed runtime for creating agents, durable threads, streamed runs, and MCP tool credential wiring in LangSmith.

Core Features & Use Cases

It helps you create and operate Managed Deep Agents (register, update, and delete) and run them on durable threads with streamed execution for long-running, observable workflows, including subagent and tool event streaming. It also supports registering and rotating MCP server credentials and configuring human-in-the-loop interrupts per tool so you can require approval before specific actions while still keeping the rest fully automated.
Common use cases include building an internet research assistant with tool-based search, orchestrating multi-step subagent workflows that need persistent state, and wiring private or credentialed MCP tools (e.g., internal endpoints) into an agent safely.

Quick Start

Start by creating a Managed Deep Agent in your LangSmith workspace, then create a durable thread for it and stream a run using the Deep Agents REST endpoints for /v1/deepagents.