Skip to main content
Use Anthropic to define reusable Claude agents, configure hosted cloud environments, run durable sessions, handle tool approvals and custom tools, inspect threads and resources, and start automations from signed lifecycle events.

Connect Managed Agents

  1. Create an Anthropic API key with access to the workspace that owns your agents.
  2. In the Anthropic Console, create a webhook with endpoint https://automate.ax/api/v1/events/anthropic.
  3. Enable the agent, environment, and session events your automations use. Enabling every supported event lets you add triggers without revisiting the webhook.
  4. Connect Anthropic in Automate.ax with Managed Agents. Enter the API key, the webhook signing key that begins with whsec_, and a workspace ID when the key can access more than one workspace.
The ordinary API key connection remains available for Anthropic model generation. Managed Agents triggers require the Managed Agents connection because it stores the signing key and workspace identity.

Run one durable turn

automations/run-claude-agent.automation.ts
anthropic.runAgent starts the session in the initiating context. The matching idle or terminated webhook starts an ordinary root context. Automate.ax correlates both by session ID and resumes in their shared child; the webhook never inherits hidden causal ancestry.

Sessions and control

Agents and environments

  • Create an agent, update it, and inspect immutable versions.
  • Create a hosted environment and control packages plus unrestricted networking or networking restricted to specified domains.
  • Session creation accepts GitHub repositories, uploaded-file references, memory-store references, vault IDs, budgets, metadata, initial messages, outcome definitions, agent overrides, MCP servers, skills, and multi-agent rosters.
Self-hosted environments, scheduled deployments, vault and credential management, memory-store management, file uploads, and skill management are outside this integration. You can attach IDs created through Anthropic APIs or Console.

All actions

All triggers

Webhook behavior

Anthropic webhooks contain resource IDs, not complete objects. Automate.ax verifies the Standard Webhooks signature and five-minute timestamp window, deduplicates retries by event ID, then fetches the authoritative agent, environment, session, and latest idle reason. Deliveries may arrive out of order and Anthropic does not provide webhook backfill, so use the resource-fetching actions when you need current state. Every action uses camelCase public fields and the official Anthropic client. Provider calls time out before the action lease. Read actions are replay-safe; creates, updates, sends, interrupts, archival, and deletion are not retried after an uncertain provider response.