Connect Claude Code to AGNT and use your own subscription
Sign in with the Claude plan you already pay for and run Claude models inside AGNT. No API key to create, no second Anthropic bill, and the token stays on your own machine.
Contents
If you already pay Anthropic for Claude, you should not need a second Anthropic bill to use Claude inside AGNT. Connect your account and your agents, workflows and chats run on the plan you already have — the same credential Claude Code itself uses, no API key required.
Four steps, and no password is ever typed into AGNT.
Before you start
You need a paid Claude plan. AGNT verifies the credential against Anthropic's API the moment you connect, so if the plan cannot serve inference you find out immediately rather than the first time a workflow fails at 3am.
What actually happens when you connect
AGNT sends you to Anthropic's own OAuth screen at claude.ai. You sign in and approve the request there, on Anthropic's domain — your Anthropic password is never entered into AGNT and never passes through it.
What comes back is an inference-scoped OAuth token, the same kind of credential the Claude Code CLI holds. AGNT writes it to the credential file on your own machine, refreshes it before it expires, and sends it straight to Anthropic when you run something. It is never uploaded to us, because AGNT runs locally and there is nowhere for it to go.
Worth stating plainly, because "connect your account" often means "hand over your password", and here it does not.
Connecting
- Open Connectors in the left sidebar
- Find Claude Code under Auth Connections and click Connect
- Your browser opens Anthropic's authorization page — sign in and approve
- Anthropic shows you a short code. Paste it back into AGNT
AGNT tests the token against the API before it reports success, so a connection that says it worked has actually worked.
If you already hold a token, the same panel accepts one pasted directly.
What it costs you
Nothing beyond the plan you are already paying for. There is no per-token charge and no separate API balance to top up.
It is still your plan, so your usage counts against your plan's normal rate limits exactly as it would in Claude Code or the Claude app. AGNT does not lift those limits and does not pretend to — if you hit a cap mid-workflow, that is the same cap you would have hit anywhere else.
When the token expires
Not in any way you will notice. AGNT refreshes it in the background before it lapses, and stores the new refresh token whenever Anthropic rotates one.
If you revoke AGNT's access from your Anthropic account, the next refresh fails as revoked. AGNT clears the stored credential and tells you to reconnect, rather than retrying a dead token or failing quietly.
Disconnecting
The same panel has a Disconnect button. It removes the Claude credential and nothing else — every other provider you have connected is untouched. Revoking from your Anthropic account works too, and AGNT will notice.
Other subscriptions you can connect the same way
Claude is not the only one. AGNT connects to the AI CLI subscriptions you are probably already paying for:
| Provider | What it signs in with |
|---|---|
| Claude Code | Your Anthropic account |
| OpenAI Codex | Your ChatGPT plan, via device login |
| Gemini CLI | Your Google account, including AI Pro and Ultra |
| Cursor CLI | Your Cursor subscription |
| Antigravity | Your Google account |
| Grok | Your xAI account |
Each connects from the same Connectors page and behaves the same way: you authorize with the provider, the credential stays local, and tokens refresh themselves. Connect several and pick a different model per agent — something cheap for routine passes, something expensive for the work that deserves it.
Getting AGNT
AGNT Community Core is free, runs entirely on your own hardware, and needs no account to use. Download it, connect Claude, and give an agent a real job.
To read how the model layer resolves providers before installing anything, see the models and providers reference.