Self-Hosted Claude Managed Agents Enter Private Beta: What You Run, What Is Missing and Who Can Join

Published:

Anthropic has opened a private beta of self-hosted Claude Managed Agents, which lets a company run the whole agent service, not just the sandboxes, in its own cloud account or data center. The beta opened on November 12, 2026 and access goes through an Anthropic account team. Anthropic’s documentation does not state a price. Here is what the customer runs, what stays inside the network, what is missing compared with the hosted service, and what to ask before you request access.

What is self-hosted Claude Managed Agents?

Claude Managed Agents runs Claude as an autonomous agent: you define an agent, start a session and exchange events while it works. Until now, a customer could host only the sandboxes where tools run commands, and Anthropic ran the rest. Anthropic’s documentation describes three options: Anthropic hosts everything (the default), the customer hosts the sandboxes, or the customer hosts the sandboxes and the service. In the third option the model runs at Anthropic or at a provider the customer chooses. Help Net Security reports the same change.

What you would run

  • Server (cma-server): serves the API, runs the agent loop and stores agents, sessions and session history. It never runs an agent’s shell or file tools.
  • Database: PostgreSQL 16 or later. Without one, the server keeps data in memory and a restart erases it.
  • Workers: programs that take work from the server and run shell and file tools in your sandboxes.
  • Console: a web page on your server where people sign in through your identity provider.

You also install each release yourself and secure the server. The documentation warns that the agent’s bash tool is a real shell with the worker’s permissions and network access, so isolation comes from the container or VM the worker runs in. Each trust level needs its own deployment.

Which models can it use?

Model requests can go to the Claude API (the default), Amazon Bedrock, Claude models on Google Cloud, Microsoft Foundry or a gateway you run. For each model call the server sends the conversation so far, including tool inputs and results, to that provider, so the model itself still runs outside your network. Help Net Security says an existing zero data retention agreement with Anthropic covers these calls like other Messages API calls. Web search works only with some providers (Claude API, Google Cloud and Foundry), and web fetch only with the Claude API and Foundry. Both are off until an admin turns them on.

What is missing compared with the hosted service?

Anthropic’s differences page says nothing is copied from the hosted service, session history included, and lists features that are not available:

  • Cloud environments, so every environment must be self-hosted.
  • The rest of the Claude API, such as the Messages and Models APIs, which return 404 on your server.
  • File session resources and the Files API.
  • MCP OAuth vault credentials; static bearer tokens are the alternative.
  • The prebuilt docx, pdf, pptx and xlsx skills.
  • Scheduled deployments, dreams and webhooks.

Some features need settings first. The server connects only to MCP hosts on an allowlist that is empty by default, and agents can push only to Git branches you list. The server does not refuse force pushes, so Anthropic advises protecting branches on your Git host.

What does it add?

Features only a self-hosted server has include single sign-on through an OpenID Connect provider, storing credentials in your own secret manager, and a signed short-lived agent identity token sent to MCP servers. Help Net Security adds that credentials stay out of the sandbox, with a proxy attaching them for MCP tools and supported API calls, and that tool calls are logged and exported over OpenTelemetry.

Price and eligibility

The documentation and Help Net Security give no price for the server software, and say access is for eligible organizations through an account team; what makes an organization eligible is not stated. The server is licensed under a separate Software License Agreement, linked from Anthropic’s documentation, which legal and security teams should read before any trial.

Questions to ask your account team

  • What counts as an eligible organization and what the beta costs.
  • Which of the unavailable hosted features you rely on today.
  • Which inference provider and region fits your data rules.
  • Who will patch the server and database, since you install each release yourself.

Sources

Image: Carl Lender / CC BY 2.0 via Wikimedia Commons.

Like our coverage? Add stepphase.com as a preferred source on Google to see more of our stories in Top Stories.

Saif Ali Tai
Saif Ali Taihttps://xplorely.com/
I'm Saif Ali Tai, a tech blogger and software engineer with 8+ years of experience building apps, and I run StepPhase as its editor. I cover technology news, AI tools and security stories that matter to everyday users. Articles here are drafted with AI assistance and checked against the sources linked in each post.

Related articles

Recent articles