AI Judge changed title after evaluation, original title: "Claude announces public beta of self-hosted sandboxes and MCP tunnels preview for Managed Agents"
Features were presented at Code with Claude London with Cloudflare and Modal support.
Boris Cherny#369
Live from Code with Claude London: we're launching self-hosted sandboxes (public beta) and MCP tunnels (research preview) in Claude Managed Agents. Run agents inside your own perimeter, with your security controls applied by default.
Many users praised Claude's self-hosted sandboxes and integrations like Cloudflare for enabling secure enterprise agent execution with their own controls, while some raised concerns about management burdens, trust, and new cost risks.
How would this work with Vercel?
Vercel Sandbox powers self-hosted execution for Claude Managed Agents by running tool calls like run_shell and read_file in fresh, isolated microVMs per session. ¹ Anthropic retains the core agent loop, memory, and state while Vercel handles a webhook-driven control plane that polls work, acknowledges it, and spawns sandboxes with credential brokering and session-scoped network policies. ¹
Setup uses a Next.js app with `@vercel/sandbox@beta`, a self-hosted Anthropic environment, and a runner script inside a prebuilt snapshot.
Can you link me to the page where I can start playing with this?
Start here: https://platform.claude.com/docs/en/managed-agents/quickstart (includes the self-hosted sandboxes section). ¹
Log into the Claude Platform console at platform.claude.com to create a self-hosted environment and run your first agent session.
it’s out!
managed agents on Cloudflare Sandboxes 🔥
Live from Code with Claude London: we're launching self-hosted sandboxes (public beta) and MCP tunnels (research preview) in Claude Managed Agents. Run agents inside your own perimeter, with your security controls applied by default.
Excited to launch Claude Managed Agents on Cloudflare today!
- Run sandboxes as microVMs or even lighter-weight isolates on CF - Zero-trust creds injection, custom egress proxies, better observability, private services via VPC - Agent Email + Browser Run + easy custom tool calls
Live from Code with Claude London: we're launching self-hosted sandboxes (public beta) and MCP tunnels (research preview) in Claude Managed Agents. Run agents inside your own perimeter, with your security controls applied by default.
Cloudflare has integrated with Anthropic's Claude Managed Agents to provide a fast, isolated execution environment for autonomous code delivery. This means builders can scale agent workflows globally while strictly controlling access to private backends and easily customizing their agent’s tools and runtimes. https://cfl.re/3R9hPA9
Modal is proud to now support @claudeai Managed Agents with Modal Sandboxes.
Live from Code with Claude London: we're launching self-hosted sandboxes (public beta) and MCP tunnels (research preview) in Claude Managed Agents. Run agents inside your own perimeter, with your security controls applied by default.
Claude Managed Agents 🤝 Vercel Sandbox https://vercel.com/kb/guide/run-claude-managed-agent-tools-with-vercel-sandbox
Live from Code with Claude London: we're launching self-hosted sandboxes (public beta) and MCP tunnels (research preview) in Claude Managed Agents. Run agents inside your own perimeter, with your security controls applied by default.
We're partnering with an up-and-coming startup called @AnthropicAI for this. I'm bullish – I think they could be big one day!
Modal is proud to now support @claudeai Managed Agents with Modal Sandboxes.
Match made in heaven @daytonaio & @claudeai
With self-hosted sandboxes, you can run agents in any environment you control: your own infrastructure, or managed providers like Cloudflare, Daytona, Modal, or Vercel.
Try self-hosted sandboxes today on the Claude Platform, and request access to MCP tunnels: http://claude.com/blog/claude-managed-agents-updates
With MCP tunnels, your agents reach MCP servers inside your private network without exposing them to the public internet.
With self-hosted sandboxes, you can run agents in any environment you control: your own infrastructure, or managed providers like Cloudflare, Daytona, Modal, or Vercel.
very excited about this launch and proud to have @cloudflare play a part in both @claudeai announcements today
looking forward to having more customers like @Amplitude_HQ build managed agents on @cloudflare + managed agents
Live from Code with Claude London: we're launching self-hosted sandboxes (public beta) and MCP tunnels (research preview) in Claude Managed Agents. Run agents inside your own perimeter, with your security controls applied by default.
Smol.
Ngl seeing @daytonaio’s logo on Anthropic’s docs is an amazing feeling
Check out our starter template here: https://github.com/cloudflare/claude-managed-agents
what better way to get the word out for the @claudeai x @modal collab than to pay @andonlabs' Claude radio station @andon_thinking to make and air an advertisement?
Live from Code with Claude London: we're launching self-hosted sandboxes (public beta) and MCP tunnels (research preview) in Claude Managed Agents. Run agents inside your own perimeter, with your security controls applied by default.
@claudeai #KeepSonnet45
Managed Agents splits the agent loop from the sandbox. @anthropic runs the loop, manages session state, handles recovery. Modal gives you the production cloud infrastructure to do this at scale, with:
– Fast startup, even on heavy custom images – Scale to 100,000+ concurrent sandboxes – Volumes, snapshots, secure networking, and the rest of Modal composable in – Access to expanded CPU and GPU resources when you need it
@claudeai
To best support its merchants, @DoorDash is building an intelligence system with a suite of AI agents. From @andyfang:
@claudeai It’s 4:30 am I literally just wake up!
@Cloudflare Fast, isolated execution at global scale, solved.
The record of what those agents did still needs somewhere durable and verifiable to live.
That's the next layer.
