Claude Code API Key vs Subscription: What the Key Turns Off

An approved ANTHROPIC_API_KEY switches off Remote Control, claude.ai connectors and /usage-credits in your local Claude Code, even while you stay signed in to a Pro or Max plan. It also hides /schedule and makes claude --cloud and claude --teleport fail, and Chrome integration, artifact publishing and voice dictation drop out with them. Anthropic’s docs tie each of those features to a claude.ai login, and an approved key outranks that login. For the Claude Code API key vs subscription choice, my first filter is that feature list; price comes second, and only for readers who need none of it.

I build this site with the Claude Code extension in VS Code on Windows 11, so the shell commands cover PowerShell and Git Bash as well as macOS and Linux.

Claude Code API Key vs Subscription: Which Should You Use?

Use your claude.ai subscription login if you rely on Remote Control, routines, cloud sessions or claude.ai connectors, or want a flat monthly fee. Use an API key for pay-per-token billing with no plan limits, occasional use, or commercial data terms without a Team or Enterprise seat. Once approved, ANTHROPIC_API_KEY outranks the login.

As of 2 October 2026:

	claude.ai subscription login	ANTHROPIC_API_KEY
Bills to	Your Pro, Max, Team or Enterprise plan	Your Claude Console organization, per token at API rates
Usage limits	Plan limits shared between Claude and Claude Code	No plan limits; you can cap spend per Console workspace
1M context on Fable, Sonnet 5 and later, and Opus 4.7 and later	Every plan, Pro included	Yes
When you run out	Wait for the reset, or turn on usage credits	Add Console credits, or turn on auto-reload
Remote Control, claude.ai connectors, /usage-credits	Available	Unavailable
Routines and cloud sessions from the CLI	Available	/schedule hidden; claude --cloud and claude --teleport fail
Chrome integration, artifacts, voice dictation	Available	Unavailable
claude --bare scripts	Can’t use the login	Need the key or an apiKeyHelper
Model training on your prompts and code	Pro and Max: your privacy setting decides; Team and Enterprise: commercial terms	Commercial terms: no training unless your organization opts in
Fits	Daily work and sessions you steer from a phone	Occasional use and scripted runs

The precedence comes from Anthropic’s authentication docs, the shared limits from its Pro and Max help article (updated 19 August 2026), 1M context from the model configuration page, and training from the data usage page. Every Claude Code fact on this page comes from Anthropic’s docs and help center as checked on 2 October 2026. The next section sources each feature row with Anthropic’s rule, and the error you see where the docs give one.

What an API Key Turns Off in Claude Code

Anthropic’s admin setup page names four of the rows below, plus Code Review, in one sentence: “Cloud sessions, Routines, Code Review, Remote Control, and the Chrome extension aren’t available through Console API keys or cloud-provider credentials alone.” Code Review is a Team and Enterprise feature in research preview, so for a Pro or Max user the list reads like this, with what each feature does once a key is in use:

Feature	With an approved API key	What Anthropic’s docs say
Remote Control	Won’t connect	”API keys are not supported.” Signed in with a key set, the error reads Remote Control requires claude.ai subscription auth. and names the cause, such as ANTHROPIC_API_KEY is set, so this session is using API-key auth
Routines and /schedule	/schedule hidden in the CLI; claude.ai/code/routines still works	”/schedule requires a claude.ai subscription login”; with a Console key, submitting it points you to Claude for Enterprise
Cloud sessions	claude --cloud and claude --teleport fail; sessions started at claude.ai/code still run on your plan	The commands fail “with Unable to get organization UUID or a message that API key authentication is not sufficient”
claude.ai connectors	Not loaded	Fetched “only when your active authentication method is a claude.ai subscription login”
/usage-credits	Unavailable	”the command isn’t available with API key authentication”
Chrome integration	Kept off, even with --chrome	”the browser extension can’t authenticate with those credentials”
Artifacts, including /design and /slides	Can’t publish	”Sessions using an API key, gateway token, or cloud-provider credential cannot publish”
Voice dictation	Off, in the CLI and in the VS Code extension	Not available “when Claude Code is configured to use an Anthropic API key directly”
Plan usage bars	Replaced by session cost	The VS Code Account & usage dialog “shows the session’s own cost and token usage instead”

The artifacts row is also why /design needs a claude.ai login, a requirement my Claude Design vs Claude Code routing guide lists next to each command.

Two of those rows would hit this site’s setup. My user settings set remoteControlAtStartup to true, so every interactive session connects to Remote Control on its own, and my agent push notifications depend on that connection; my settings map for Claude Code in VS Code lists both settings. My Semrush keyword data reaches Claude Code as a claude.ai connector, listed in the setup table of my Claude vs Claude Code comparison. With a key approved, Remote Control, those pushes and the Semrush tools would all stop. For a session where Remote Control is unavailable, my guide to Claude Code alerts at the desk and away from it lists the routes to a phone that Anthropic documents, next to the two push toggles.

Other Differences Between a Key and a Login

Behavior	claude.ai subscription login	ANTHROPIC_API_KEY
claude -p runs	Used when no key is set	”the key is always used when present”, even a key you declined in an interactive session
claude --bare	Never read: bare mode “doesn’t use your subscription login”	Read, and so is an apiKeyHelper passed in the --settings JSON
Prompt cache for the main conversation	One hour by default, within plan usage	Five minutes by default; setting promptCacheTtl to 1h (v2.1.242 or later) changes it
The older Opus 4.6 and Sonnet 4.6 [1m] variants	Need usage credits: both on Pro, Sonnet 4.6 on every plan	Full access

Sources: bare mode, cache lifetime, and, linked above, the authentication docs for -p and the model configuration page for the [1m] variants. If you script Claude Code on a subscription, watch the headless page: it says --bare “will become the default for -p in a future release”.

Claude Code Authentication Precedence: The Seven Credentials in Order

The precedence section of the authentication docs opens with “When multiple credentials are present, Claude Code chooses one in this order”, and the table below adds what each one bills to and what it loses, as of 2 October 2026:

Rank	Credential	How it gets set	Bills to	Unavailable with it
1	Cloud provider	CLAUDE_CODE_USE_BEDROCK, CLAUDE_CODE_USE_VERTEX or CLAUDE_CODE_USE_FOUNDRY	Your cloud account, per token	Everything in the feature table above
2	ANTHROPIC_AUTH_TOKEN	An environment variable, sent as a bearer token to an LLM gateway or proxy	Depends on the gateway or proxy	Everything in the feature table above
3	ANTHROPIC_API_KEY	An environment variable; interactive sessions ask you to approve it once	Your Console organization, per token	Everything in the feature table above
4	apiKeyHelper	A settings key that runs a script printing a key	The account behind that key	Everything in the feature table above
5	CLAUDE_CODE_OAUTH_TOKEN	A one-year token from claude setup-token, for CI and scripts	Your Pro, Max, Team or Enterprise plan	Remote Control, claude.ai connectors, Chrome; bare mode ignores it
6	Anthropic profile or federation credentials	ANTHROPIC_PROFILE, the federation variables, or an active profile set up for Workload Identity Federation	Your Console organization	Everything in the feature table above
7	Subscription login	/login with your claude.ai account	Your plan’s usage limits	Bare mode only

Every row of the feature table needs a claude.ai login, and a credential at ranks 1 to 6 is chosen ahead of that login, so ranks 1 to 4 and 6 lose every one of those features. For ANTHROPIC_AUTH_TOKEN, Anthropic’s gateway overview says setting it “turns off subscription login for that session”. Rank 5 bills to your plan: the setup-token “authenticates with your Claude subscription”, per the long-lived token section, but it “can only make model requests”, which rules out Remote Control and claude.ai connectors, and the Chrome page rules out Chrome as well. The other feature pages don’t say how it fares on their rows. The docs add these rules around the order:

• Interactive sessions ask once: “you are prompted once to approve or decline the key, and your choice is remembered.” The “Use custom API key” toggle in /config changes that choice later, and it only appears while the variable is set.
• The keyless Console sign-in, from v2.1.242, stores an Anthropic profile instead of a key, and it “signs you out of any claude.ai login stored on the machine”. Its profile ranks at 6 only when ANTHROPIC_PROFILE names it; otherwise it ranks below a working /login.
• Cloud sessions sit outside the order once they run: they “always use your subscription credentials”, and a key set in the cloud environment “doesn’t override” them. A key on your own machine still stops claude --cloud and claude --teleport from starting or pulling one.
• A signed-in Claude apps gateway session, which an organization self-hosts, also sits outside the list and outranks all seven.

Which Credential Is Claude Code Using Right Now?

Run /status: a Login method row such as Claude Max account means your subscription is in use, and an API key row means “your requests bill to a Claude Console organization”, per Anthropic’s note on where fast mode spend appears. With a login and a key both configured, /status marks the one that isn’t in use.

Where	Run	What it tells you
Claude Code CLI	/status	A Login method, API key or Profile row
VS Code extension	/status, or Status in the Customize section	Version, account, model and MCP servers, on v2.1.280 or later
Any terminal, outside a session	claude auth status --text	Your sign-in state; the JSON form’s authMethod reads none, claude.ai, oauth_token, api_key, api_key_helper or third_party
A shell	echo $env:ANTHROPIC_API_KEY in PowerShell, echo %ANTHROPIC_API_KEY% in Command Prompt, echo $ANTHROPIC_API_KEY in Git Bash, macOS or Linux	Whether the variable is set in that shell

Sources: the VS Code guide, the CLI reference, and Anthropic’s help article on managing API key environment variables, updated 5 May 2026. The claude commands need the standalone CLI, since the VS Code extension doesn’t put claude on your PATH.

These messages point at a key that has taken over:

• Credit balance is too low on a Pro, Max, Team or Enterprise plan, where the error reference says to “run /status and check the API key row”
• Your ANTHROPIC_API_KEY belongs to a disabled organization, from an old key that still overrides your login
• A connector you added at claude.ai missing from /mcp, which the MCP docs say to check with /status

How to Make Claude Code Use Your Subscription Instead of an API Key

1. Decline the key in interactive sessions. In the CLI, turn off “Use custom API key” in /config. Runs with claude -p still use the key.
2. Remove it from the current shell. Run unset ANTHROPIC_API_KEY in Git Bash, macOS or Linux, Remove-Item Env:ANTHROPIC_API_KEY in PowerShell, or set ANTHROPIC_API_KEY= in Command Prompt.
3. Remove it where it is set for good. Check your shell profile (~/.bashrc, ~/.bash_profile or ~/.zshrc, or in PowerShell the profile script whose path $PROFILE holds), the Windows Environment Variables dialog under both User variables and System variables, the env block of each Claude Code settings file, any apiKeyHelper setting, and the VS Code extension’s environmentVariables setting.
4. Restart the terminal and VS Code. Anthropic’s environment-variable article lists “Not restarting your terminal after changing environment variables” as a common issue, and its VS Code guide notes that VS Code may not inherit a shell’s environment unless you launch it from that shell with code ..
5. Run /login and choose your claude.ai account. If you signed in to Claude Code with a Console account, the Pro and Max help article says /login switches you to your plan. An approved key in your environment takes precedence over the login, per the 401 error entry, so /login alone won’t move you off it.
6. Confirm with /status. The Login method row should name your plan, and an API key row, if one still shows, should be marked as not in use.

To stop Claude Code offering API credits when you reach a plan limit, the same Pro and Max article says to log out and log back in with only your plan credentials: “Avoid adding any Claude Console credentials during the login process.”

How to Log In to Claude Code With an API Key

• A Claude plan doesn’t come with API access. Anthropic’s help center says a paid subscription “doesn’t include access to the Claude API or Console”, so a key means a separate Console account.
• For one session, set the key in that shell only and close it afterwards: $env:ANTHROPIC_API_KEY="your-api-key-here" in PowerShell, or export ANTHROPIC_API_KEY='your-api-key-here' in bash.
• To pay through the Console without storing a key, use the keyless sign-in: at the /login prompt, pick the Anthropic Console account, then “Sign in with your Console account”.
• claude auth login --console also signs in from a terminal “for API usage billing instead of a Claude subscription”. The CLI reference doesn’t say whether that stores a key or a profile, so check /status afterwards for an API key or Profile row.
• To keep a work Console account apart from a personal plan, give each its own CLAUDE_CONFIG_DIR. Each directory keeps “its own settings, session history, and claude.ai login or API key”, so in the work directory choose “Create an API key” at the Console sign-in, which stores the key with that directory’s credentials. Don’t use the keyless sign-in for the work side: separate directories don’t isolate it, because Claude Code stores it outside the configuration directory, and it would sign you out of the personal claude.ai login.
• To run CI or cron jobs on your plan, generate a token with claude setup-token and set it as CLAUDE_CODE_OAUTH_TOKEN.
• To cap a scripted run, pass --max-budget-usd, which stops a print-mode run at a dollar amount.

Claude Code Data Terms: Consumer Plans vs Commercial Accounts

Anthropic’s data training policy splits accounts into consumer and commercial, and the line doesn’t fall between a subscription and a key: a Team seat signed in with /login sits on the commercial side with the API.

Account	Training on Claude Code prompts and code	Retention
Free, Pro, Max (consumer)	Used to train new models “when this setting is on (including when you use Claude Code from these accounts)“	5 years with model improvement allowed, 30 days without
Team, Enterprise (commercial)	Not used, unless your organization opts in	30 days standard; zero data retention for qualified Enterprise accounts
API key through the Console (commercial)	Not used, unless your organization opts in, for example through the Development Partner Program, which covers first-party API use only	30 days standard

On Pro and Max, /privacy-settings views and updates that setting inside Claude Code, which the commands reference lists as “Only available for Pro and Max plan subscribers”, and claude.ai/settings/data-privacy-controls holds the same control on the web. If client briefs or unreleased positioning pass through your sessions on Pro or Max, check that setting first. An API key moves the same work under commercial terms without a Team plan.

Claude Code API vs Subscription Cost: The $13 Break-Even

Anthropic’s costs page gives one benchmark: “Across enterprise deployments, the average cost is around $13 per developer per active day and $150-250 per developer per month, with costs remaining below $30 per active day for 90% of users.” The same page says per-developer costs “vary widely based on model selection, codebase size, and usage patterns”, so a marketer’s own sessions are a better guide. The rules I would use:

• Take your plan’s monthly price from the plan table in my Claude vs Claude Code comparison linked above, or from Anthropic’s pricing page, and divide it by $13. The result is the number of active days a month at which a key, at that average, costs the same as the plan.
• If you pay from India, that monthly price depends on where you check out: I compare what each checkout lists for a Claude plan in a separate guide, and a rupee price needs converting to dollars before you divide.
• Check your own number. On a subscription, the Session block of /usage in the CLI computes a dollar figure “locally from token counts at list price”; Anthropic says the figure “isn’t relevant for billing purposes” on Pro and Max. The VS Code extension’s Account & usage dialog shows plan bars instead on a claude.ai plan, so run claude in VS Code’s integrated terminal, with the standalone CLI installed, to see the figure. /clear resets it, so read it before you clear.
• Allow for the cache. With the lifetimes in the table above, a break of five to 60 minutes “reprocesses your full context” on a key but not on the plan. The caching page linked above adds that the hour-long cache “bills cache writes at a higher rate” and “costs more on short bursts of work that never idle past five minutes”, so a subscription session’s list-price figure can land above or below what the same work costs on a key.
• Usage credits are “billed at standard API rates”, per Anthropic’s usage credits article, so a month that runs mostly on credits is the month to price a key against the plan. The four ways back from a Claude usage limit cover credits and resets in full.
• Per-token prices for each model, and what common marketing jobs cost on them, are in my Claude Sonnet vs Opus vs Haiku cost guide.

Subscription or API Key: A Verdict by Usage Pattern

If you	Use	Why
Work in Claude Code most weekdays	Subscription login	A flat fee covers daily work within plan limits, and Remote Control and connectors stay on
Open it a few days a month	An API key, once the break-even check agrees	You pay only for active days
Want routines or cloud sessions	Subscription login	Both need a claude.ai account; my routines vs scheduled tasks guide compares the schedulers
Run scheduled claude -p scripts	CLAUDE_CODE_OAUTH_TOKEN on your plan, or a key with --bare and --max-budget-usd	The token keeps scripts on the plan; --max-budget-usd caps what a key can spend
Put client material through Claude Code on Pro or Max	Your privacy setting first, or a key for that work	API use runs under commercial terms
Hold a work Console account and a personal plan	A separate CLAUDE_CONFIG_DIR with a work API key, not the keyless sign-in	A key exported in your shell profile would sit above the personal login everywhere

Claude Code API Key vs Subscription: Rule Out Features First

If you use any feature an API key turns off, keep the subscription login; the $13 break-even check only matters for the rest.

Open your next session with /status and read which row it shows.