Claude Managed Agents + N8N: Full Production Tutorial

What you can actually set up from this

Extracted from the video's own transcript — the specifics the original summary left out.

Configuration shown

Create-session request body — the three IDs that matter

{
  "agent_id":       "<which agent — e.g. the customer support bot>",
  "environment_id": "<which container, with its tools and network policy>",
  "vault_ids":      ["<credential grant, e.g. the ClickUp OAuth>"]
}

Template shown in the video. Adapt to your project — do not copy verbatim without reviewing each line.

Reproducible steps

1. Create the agent in the Claude console

   Go to platform.claude.com and add credits. You can start from a template, from a clean slate, or by describing what you want to the assistant — e.g. a customer-support bot for one domain that answers FAQs, troubleshoots, and raises a task in ClickUp when it cannot fix the problem or the customer asks for escalation.

2. Configure the environment

   The environment is the cloud container configuration the agent runs inside — the room it works in, with its tools and files. Set network access here; the walkthrough restricts web search to a single domain rather than leaving it open.

3. Authorise tools in the credentials vault

   The session will not start until outstanding MCP authentication is done. ClickUp is connected via OAuth, after which it appears in the credentials vault and can be referenced by ID.

4. Test in the console before wiring anything up

   The right-hand testing panel accumulates the session; the left shows a play-by-play. Use the debug section to confirm a tool call actually fired and returned success, then verify the side effect in the target system (the ClickUp task really appears).

5. Map n8n sessions to Claude sessions

   The database layer holds exactly two fields: the n8n user session ID and the Claude session ID. On each inbound message, look up whether that user session already has a Claude session. If not, create one; if so, append to it. This is what keeps each customer's context private to that customer.

6. Create session, send message, wait, list messages

   Four API calls. Create a session (JSON body needs the agent ID, the environment ID, and the vault IDs for any authorised tools), send the message in, wait for Claude to execute tool calls and generate, then list the session's messages and take the most recent agent message to display. The whole integration is about ten nodes.

7. Deploy

   Publish the workflow, open the chat node, copy its URL and use it in a browser. Only the create-session call needs editing per agent — everything downstream carries through from the stored session ID.

Gotchas

• Managed agents do NOT run on your Pro or Max subscription. They bill against API tokens on the account's credits, which is why you must add credits before starting.
• There is a second charge on top of tokens: $0.08 per session hour, billed only while the agent is actually running. Idle time between messages is free — a 2 minute 28 second gap in the walkthrough is shown as idle and uncharged.
• The default model at the time of recording is Sonnet 4.6, and the video notes this may change — check it rather than assuming.
• The 15-second wait node is explicitly called out by the author as NOT production ready. Real tasks can take minutes, so a fixed sleep will silently truncate them. Replace it with a polling loop before shipping.
• Prefer one environment per agent (separation of concerns) rather than sharing one environment across several agents.
• Sessions are the unit of isolation: one agent, many sessions — one per customer, per invoice, or per day. Reuse the agent, never the session.