Claude Code
Anthropic's coding agent, in your terminal.
Agents
One command installs skills, MCP, and a least-privilege credential.
Aptarium detects the coding agent already on the machine, wires it up, and hands it a scoped token — the secret is installed but never printed. That token gives the agent governed, read-as-the-viewer access to the ten connectors live at launch, so it can build and publish a readout from the tools you already run.
Codex
OpenAI's CLI coding agent.
Cursor
The AI-native editor.
Copilot
GitHub's assistant, where you already write code.
Pairing
Detect, install, verify, record.
Setup signs in with Google, detects installed agents, installs skills and MCP configuration, mints a 90-day credential, verifies MCP initialization, and records the device pairing. From there the agent can deploy and publish readouts without ever handling a connector key. A failed local write rolls everything back and revokes the token.
$ aptarium setup acme
✓ paired · claude-code scopes data:read apps:read apps:deploy
$ aptarium setup acme --agent codex
✓ paired · codex credential rotated in place
Scopes
Least privilege by default. Sharing is an explicit opt-in.
ScopeGranted by default
data:readDefault
apps:readDefault
apps:deployDefault
share:manageOnly with --share
Rotatable
Rerun setup to replace the device credential in place.
Recoverable
A failed configuration is restored and its token revoked.