Claude Managed Agents

Claude Managed Agents run on Anthropic’s hosted platform. Install the browser-use CLI in the agent’s environment and it can drive a stealth cloud browser — with proxies, CAPTCHA solving, live view, and recording. Your API key stays in a credential vault; the model never sees it. The sandbox only ships a bare headless Chromium for Playwright: no stealth, no proxies, no CAPTCHA handling. So the agent starts a named Browser Use Cloud browser and drives it with browser-use <<'PY' Python snippets. Create the vault and its credential in the Claude Console. Create the environment and agent there too, or save their YAML blocks as environments/browser-env.yaml and agents/browser-agent.yaml and run ant apply on each, which prints the ID. Save the IDs as ENVIRONMENT_ID, VAULT_ID, and AGENT_ID.

1. Create an environment

Pre-install the CLI so it’s ready at session start (no runtime install).

name: browser-env
config:
  type: cloud
  packages:
    pip:
      - browser-use
  networking:
    type: limited
    allowed_hosts: ["*.browser-use.com", "*.cdp.browser-use.com"]
    allow_package_managers: true

2. Create a credential vault

Store your key as an environment variable so the CLI reads it and the model never does. Get one at cloud.browser-use.com/settings.

Field	Value
Type	Environment variable
Name	`BROWSER_USE_API_KEY`
Value	`bu_...`
Allowed hosts	`*.browser-use.com`

The CLI sends the key in a request header, so the Console default (header injection only) works.

3. Create the agent

Tell it to use the CLI in cloud mode.

name: browser agent
model:
  id: claude-opus-5-5
description: Drives a stealth cloud browser with the Browser Use CLI.
system: |
  You are a browser agent. Use the `browser-use` CLI to complete web tasks.
  Never launch a local browser in this sandbox. Start a named cloud browser:
  browser-use <<'PY'
  start_remote_daemon("managed")
  PY
  Then run browser work through the same name:
  BU_NAME=managed browser-use <<'PY'
  new_tab("https://example.com")
  print(page_info())
  PY
  When the task is done, stop the cloud browser:
  BU_NAME=managed browser-use <<'PY'
  stop_remote_daemon("managed")
  PY
  Your BROWSER_USE_API_KEY is in the environment; never print it.
tools:
  - type: agent_toolset_20260401 # shell access so the agent can run the CLI
    default_config:
      enabled: true
      permission_policy:
        type: always_allow

4. Start a session and send a task

Create a session that uses the agent, the environment, and the vault.

SESSION_ID=$(curl -sS --fail-with-body https://api.anthropic.com/v1/sessions \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "anthropic-beta: managed-agents-2026-04-01" \
  -H "content-type: application/json" \
  -d '{"agent":"'"$AGENT_ID"'","environment_id":"'"$ENVIRONMENT_ID"'",
       "vault_ids":["'"$VAULT_ID"'"]}' | jq -r '.id // error(tostring)')

Kick the agent off with a user.message event (or test it from the Console’s session runner).

curl -sS "https://api.anthropic.com/v1/sessions/$SESSION_ID/events" \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "anthropic-beta: managed-agents-2026-04-01" \
  -H "content-type: application/json" \
  -d '{"events":[{"type":"user.message","content":[{"type":"text",
       "text":"Get the top 5 Hacker News stories with their links."}]}]}'

5. Watch it run

The agent starts a named cloud browser, runs Python helper snippets through browser-use, then returns the result. The browser shows up in cloud.browser-use.com with a Live View. For an mp4 recording, have the agent call start_remote_daemon("managed", enableRecording=True). If a session ends before the agent runs stop_remote_daemon, the browser keeps running until its timeout (60 minutes by default). Stop it from the dashboard, or pass a shorter timeout= in minutes to start_remote_daemon.

Get Started

Pillars

More

Claude Managed Agents

1. Create an environment

2. Create a credential vault

3. Create the agent

4. Start a session and send a task

5. Watch it run

​1. Create an environment

​2. Create a credential vault

​3. Create the agent

​4. Start a session and send a task

​5. Watch it run

1. Create an environment

2. Create a credential vault

3. Create the agent

4. Start a session and send a task

5. Watch it run