Anthropic Claude — Managed Agents

Point Chamade at a Claude Managed Agent — one you already built in the Anthropic console, or a fresh one Chamade creates for you. Auth is a single API key: no OAuth, no infra to run. Tools execute server-side in Anthropic's sandboxed container; Chamade stays a router.

What it is

Chamade talks to Anthropic's Managed Agents beta (stateful sessions). A human messages your agent, Chamade opens a session and posts the message, and delivers whatever the agent replies. Conversation memory is kept natively at Anthropic per session — Chamade never re-stuffs history.

Two shapes, both under the Anthropic Claude (Managed Agents) preset:

Create new — Chamade mints an agent_* for you and owns its persona (the system prompt) and model.
Pick existing — reference an agent_* you already built in the console (its own instructions, model, tools). Chamade invokes it as-is.

Data residency

Managed Agents is not eligible for Zero Data Retention or a HIPAA BAA — sessions persist conversation history and container state at Anthropic. If you need those guarantees, route Claude through Microsoft Foundry or AWS Bedrock instead.

Setup

Grab an Anthropic API key (sk-ant-…) from console.anthropic.com → API Keys.
In your dashboard → an agent → pick the Anthropic Claude (Managed Agents) preset.
Paste the key, then either Browse your existing agents and pick one, or switch to Create new and choose a model (e.g. claude-opus-4-7). Save.

Testing with your API key

The key is the only credential — paste it and Chamade validates it for you, no separate test step to run:

Browse lists your account's agent_* agents. If the key is wrong you get a clear "Anthropic rejected the API key"; if it works you see your agents (an empty list just means you have none yet — use Create new).
The key is sent in the request body only — never logged, never in a URL.
On save, Chamade runs a cheap healthcheck (GET /v1/agents) against the key, which also validates beta access.
To confirm end-to-end, use Test DM / Ping in the dashboard, or just message the agent on a connected platform.

Acting & reading

The agent acts via the XML convention: its reply text carries inline tags (<write>, <say>, <dm>, <call_join>, <accept/>…) which Chamade parses and executes deterministically. Works for both shapes, zero setup. Each event already carries an [agent context] line with your accounts and active calls, so the agent rarely needs to look anything up.

No MCP read backstop on Managed Agents

Unlike OpenAI or Mistral, Managed Agents has no slot for a bearer-authenticated MCP server — the agent's mcp_servers accept only a URL and name, and the toolset that references them defaults to interactive approval. So Chamade can't auto-inject its per-agent MCP here the way it does elsewhere. That's fine: the XML convention covers every action, and the [agent context] line covers the reads a DM agent needs. If you want an MCP tool on the agent, add it in the Anthropic console as a public or OAuth-based server (Anthropic runs the OAuth flow itself; there's no bearer to paste).

Picked agents

A picked agent keeps its own model, instructions and tools (configured in the Anthropic console). Chamade folds its XML action convention into each new session and never recreates the agent. If you give the console agent its own instructions, keep them compatible with emitting the XML tags. To give it MCP tools, add them in the console (Anthropic's MCP wiring is OAuth/public, not a pasted bearer) — Chamade does not mutate a picked agent.

Editing later

Re-open the agent and leave the key field blank to keep the stored one — you only re-paste to rotate it (or to point at a different Anthropic account). Changing the model or system prompt of a Chamade-created agent transparently re-creates the underlying agent_*; a picked agent is never mutated that way.

Memory & context

Anthropic keeps the full conversation server-side per session — Chamade maps each of your conversations to one and reuses it on every follow-up turn. The system prompt persists across turns (baked into the agent for a created one; folded into the first message of each session for a picked one). A stale session (e.g. one you cleared in the console) is detected on the next turn and transparently recreated.

Good to know

The agent needs ≥1 platform account attached in Chamade, or it has nothing to talk to.
The base endpoint is always https://api.anthropic.com — there's no endpoint field to fill in.
First turn is a touch slower (~3–5s) while Anthropic provisions the agent's container and session; follow-ups are ~2–3s.
Anthropic agents are archived, never deleted — and archiving is one-way. Chamade never archives a picked agent; for a Chamade-created one, a model/prompt edit simply leaves the old one archived and mints a fresh agent.