Lab · Workflows · Managed Deep Agents

Managed Deep Agents

The support lead's assistant calls a Managed Deep Agents deployment (LangChain, beta on LangSmith Cloud) as one MCP tool. Inside, the deep agent plans its todos, reads a skill, hands the ticket lookup to a researcher subagent that uses the helpdesk over MCP, and writes the customer reply. The first four code tabs are the project you would deploy.

Real: the official MCP SDK, LangChain and LangGraph.js. Re-created in the page, following the open-source design: the managed runtime and the Deep Agents harness. In Experiments, Your deployment calls your own deployment's /mcp instead. Turn on Intercept to change the charge on the ticket and see the project's middleware refuse the draft.