# Run Claude Managed Agents with Chat SDK

Source: [Vercel](https://vercel.com/changelog/claude-managed-agents-with-chat-sdk)  
Feed7 permalink: https://feed7.dev/p/claude-managed-agents-with-chat-sdk-0wtme2i  
Published: 2026-07-27T00:00:00.000Z  
Trust: Official Source (official_source)

## Why Included

Chat SDK can put Claude Managed Agents behind browser and messaging interfaces while Anthropic retains the loop, tools, session state, and research sandbox.

## Source Summary

Claude Managed Agents now connects to **Chat SDK** through one type-safe handler. The managed service owns the model loop, tools, session state, and sandboxed web research; the SDK adds streaming replies and a live activity feed.

## Practical Implication

Use the managed session as the conversation source for transcripts, replay, and sidebars instead of operating a separate state database. Adapters can move the same handler to Slack, Teams, Discord, WhatsApp, and **30+ platforms**.

## Agent-Ready Context

Claude Managed Agents now connects to **Chat SDK** through one type-safe handler. The managed service owns the model loop, tools, session state, and sandboxed web research; the SDK adds streaming replies and a live activity feed.

Use the managed session as the conversation source for transcripts, replay, and sidebars instead of operating a separate state database. Adapters can move the same handler to Slack, Teams, Discord, WhatsApp, and **30+ platforms**.

This is a hosted architecture rather than a portable agent runtime: key execution and state responsibilities remain with Claude Managed Agents. The browser quickstart still requires **Anthropic credentials**.

## Connected Context

Feed7 judgment across 262 accumulated Signals:

This collapses agent execution, tools, session history, streaming UI, and channel delivery into a hosted path, reducing the application state and adapter work teams must own. It also sharpens the portability tradeoff: transcripts and replay can use the managed session directly, but execution and state remain tied to Claude Managed Agents and still require Anthropic credentials.

- [A Song of Types and Agents - Roberto Stagi, Ratel](https://feed7.dev/p/a-song-of-types-and-agents-roberto-stagi-ratel-0i7wvpu) — The shared-types approach reinforces the value of the type-safe handler, while this integration goes further by placing the loop and session state behind a managed service boundary.
- [Gemini API Managed Agents: 3.6 Flash, hooks, and more](https://feed7.dev/p/expanding-managed-agents-gemini-api-3-6-flash-hooks-0xce1pm) — Gemini’s hooks, caps, schedules, and model selection expose controls for unattended work that are not described here, highlighting different control surfaces within the same hosted-agent architecture.
- [Expanding Managed Agents in Gemini API: background tasks, remote MCP and more](https://feed7.dev/p/expanding-managed-agents-gemini-api-1idue29) — Gemini’s background tasks, remote MCP, client functions, and credential rotation broaden managed execution capabilities; the Claude integration instead emphasizes session-backed chat UI and cross-channel adapters.

## Context Map

- Layer: tools
- Domains: research
- Topics: cloud-agents, agent-sdks, dev-ux

## Uncertainty

- This is a hosted architecture rather than a portable agent runtime: key execution and state responsibilities remain with Claude Managed Agents. The browser quickstart still requires **Anthropic credentials**.

## Agent Instruction

Use this item as source-backed context. Do not invent claims beyond the linked source. If this item conflicts with another source, call out the conflict.