Introduction

Registering an agent built on the Anthropic Agent SDK brings its runs under governance the same way any other framework does. The one thing to get right first is which of two similarly named entries you want — they do quite different jobs.

Two entries, two different things

	Anthropic Agent SDK	Anthropic
What it registers	An agent you built	Agents hosted on the platform
How assets arrive	You register, then instrument	A discovery sweep finds them
Found in	`Register Agent` only	`Add Application`, `Register Agent`, `Add Model`
Needs instrumentation?	Yes — you wire the exporter	No — the connection collects

Note: If you wrote the agent, this is the entry you want. If the agents already exist on a platform and you want an inventory of them, you want the connection instead — see Anthropic (Managed Agents). Registering the same agent through both routes produces two records and splits its history.

The flow

Register the agent — Platform, name, owner. It arrives as Setup Pending — registered and not yet sending.
Open its Setup tab — It shows "Connect <platform>" — "Instrument the agent to bring its traces in." — with a Waiting for traces… pill and numbered steps: install dependencies, point the agent at your UnifyApps endpoint (UNIFYAPPS_COLLECTOR_ENDPOINT, UNIFYAPPS_API_KEY), add credentials, initialise tracing, group traces into a session, and run the example.
Follow what the panel shows — Setup content is maintained per platform, so the panel is the source rather than anything written down elsewhere — including this page.
Run the agent once — Status flips to Active when the first trace lands. The page polls, so no refresh is needed.
Point a policy at it — Instrumentation gives observation. Governance needs coverage — see Apply a Policy to Assets.

What to govern on an SDK-built agent

An agent you wrote almost certainly has tools, and tools are where the consequential risks are. In the order that earns its keep fastest:

Blocked tools on Tool name — Deterministic, costs nothing per request, and prevents rather than reports.
Secrets and PII on Tool input — The exfiltration surface — data leaving your boundary in a tool argument.
Prompt injection on Tool output — Instructions hidden in content the agent fetched itself. Harder to think of than the user-typed kind, and more likely to work.

See Where Controls Run.

One limit to be clear about

Anthropic Agent SDK agents can be registered as Observe · Advisory or Control · Enforced. Choose Control · Enforced only if the agent routes its calls through UnifyApps; on Observe · Advisory controls evaluate and record, and only Flag takes effect.

So prevention lives inside the agent. Do not give it a tool it must never call. See Governing Vendors Without a Control API.

Troubleshooting

Symptom	Check
Stuck on `Setup Pending`	Has the agent run since instrumentation? Can the exporter reach the endpoint from where it runs?
Two records for one agent	Registered here and found by the managed-agents connection. Keep one
Model calls captured, tool calls missing	Only the model client is instrumented. Tool spans need the SDK's own tracing — and without them no tool-surface control can fire
Controls never fire	Check the agent's `Policies` tab. Observed and ungoverned looks identical to well-behaved