Third-Party Apps Now Draw From Your Extra Usage, Not Your Plan Limits
Third-Party Apps Now Draw From Your Extra Usage, Not Your Plan Limits
If you opened Cursor, Claude Code, or another AI tool today and saw the message "third-party apps now draw from your extra usage, not your plan limits," here is what actually changed, why it matters, and what you need to do about it.
What This Message Means
Anthropic split Claude billing into separate pools. Your Pro or Team subscription still covers direct usage on claude.ai and the mobile app. But third-party apps that connect to Claude through OAuth (not API keys) now pull from a different balance called "extra usage credits."
This is not a bug. It is a deliberate billing change. The message appears when a third-party app tries to use Claude on your behalf and your extra usage credit balance is zero.
Which Apps Are Affected
The change applies to any tool that authenticates through your Claude account rather than using its own API key. Here is the full breakdown:
| App category | Examples | Draws from extra usage? | |---|---|---| | Code editors with Claude integration | Cursor, Windsurf, Cline, Continue | Yes | | Claude's own CLI tools | Claude Code, Claude for Terminal | Yes | | VS Code extensions | Claude Dev, Anthropic extensions | Yes | | MCP-connected tools | Any tool using Model Context Protocol through your account | Yes | | claude.ai web and mobile | Browser chat, iOS/Android app | No (uses plan limits) | | Direct API calls | Your own scripts using an API key | No (uses API billing) |
Key distinction
If the tool asks you to sign in with your Claude account, it uses extra usage credits. If the tool asks for an API key, it uses API billing. These are two completely separate systems.
How the Billing Split Works
Before this change, everything ran through one pool. Now there are three:
Pool 1: Plan limits
Your Pro ($20/month) or Team ($30/seat/month) subscription includes a set amount of usage on claude.ai and the mobile app. When you hit the cap, you get throttled to slower responses until it resets. This pool has not changed.
Pool 2: Extra usage credits (new)
This is where third-party apps now draw from. It is a prepaid balance. You load it up manually or set auto-reload. When it drops to zero, every third-party app stops working and shows the "draw from your extra usage" message. No waiting period will fix this; you have to add credits.
Pool 3: API billing
If you use an Anthropic API key directly (for your own scripts, bots, or automations), that is billed separately on a pay-per-token basis. This has always been separate and is not affected.
What Credits You Get to Start
Anthropic added a one-time starter credit to ease the transition. The amount varies by plan:
| Plan | Starter credit | Recurring? | |---|---|---| | Free | $0 | No | | Pro ($20/month) | $20 | No, one-time only | | Team ($30/seat/month) | $100 | No, one-time only | | Enterprise | $200 | No, one-time only |
To claim it, go to claude.ai/settings/usage and look for the credit banner. If you do not see it, your account may have been credited automatically. Check your extra usage balance.
Important
The starter credit is a one-time transition bonus. After you use it up, you need to add more credits manually or enable auto-reload. There is no second free credit.
How to Set Up Auto-Reload
If you use third-party tools daily, you will want auto-reload so you never get interrupted mid-session:
- Go to claude.ai/settings/usage
- Find the "Auto-reload credits" section
- Choose a reload amount ($5, $10, or $20)
- Set a trigger threshold (e.g., reload when balance drops below $5)
- Add a payment method if you have not already
How Quickly Do Credits Get Used?
Usage varies dramatically depending on your tools and intensity:
| Tool and mode | Approximate cost per hour | $20 credit lasts | |---|---|---| | VS Code extension, occasional questions | $0.10 - $0.30/hr | 65+ hours | | Cursor with autocomplete and chat | $0.50 - $1.50/hr | 13 - 40 hours | | Claude Code, normal single-agent use | $1 - $3/hr | 7 - 20 hours | | Cursor heavy agent mode | $2 - $5/hr | 4 - 10 hours | | Claude Code with parallel agents | $5 - $15/hr | 1 - 4 hours |
The biggest factor is whether your tool runs in "agent mode" (where it makes many sequential API calls) versus simple chat or autocomplete (single-shot calls). Agent mode burns through credits much faster because each step in the agent loop is a separate API call.
If Your App Uses an API Key Instead
Some tools let you bring your own Anthropic API key rather than authenticating through your Claude account. If your tool supports this, switching to an API key sidesteps the extra usage system entirely. You pay per token through standard API billing instead.
Tools that support custom API keys:
- Cursor: Settings, Models, add Anthropic API key
- Continue (VS Code): config.json, set
apiKeyfield - Cline: Settings, API Provider, enter key
- Claude Code: set
ANTHROPIC_API_KEYenvironment variable
With an API key, you get the same models and capabilities, but billing goes through your API account. This can be cheaper or more expensive depending on your usage patterns, but it gives you direct control and avoids the credit system.
Frequently Asked Questions
Is this a rate limit or temporary throttle?
No. It is a hard stop. Your requests will keep getting rejected until you add credits. Waiting will not fix it. The only way to resume is to add extra usage credits or switch to an API key.
Does this affect my claude.ai usage?
No. Direct usage on claude.ai and the Claude mobile app still draws from your plan limits. The separation only affects third-party apps.
Can I use my plan's included usage for third-party apps?
No. The pools are completely separate. Even if you have unused plan capacity on claude.ai, your third-party apps will be rejected if your extra usage balance is zero.
What happens when the starter credit runs out?
You need to add more credits manually at claude.ai/settings/usage or enable auto-reload. There is no second free credit and no grace period.
Is there a way to see which app used how much?
Currently, the usage page at claude.ai/settings/usage shows your total extra usage balance and spending history. Individual per-app breakdowns may be available depending on your plan.
Quick Fix Checklist
The billing change is permanent. Plan your credit management accordingly, especially if third-party AI tools are part of your daily workflow.
Fazm is an open source macOS AI agent. Open source on GitHub.