[FEATURE] Bundled Daily API Allotment for Subscription Plans to Enable Sanctioned Third-Party Agentic Platform Access

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Summary

Add a daily API token allotment to Pro and Max subscription plans, sized to reflect realistic human usage patterns, that subscribers can use with third-party agentic platforms. Critically, this allotment would be drawn from a single unified pool shared across all access surfaces — front-end consumer interfaces and API usage alike — so that consumption through any channel counts against the same underlying budget. Allotments reset on a rolling window, and once exhausted, users are throttled until reset or can purchase additional tokens.

Background & Motivation

Anthropic has recently and understandably tightened enforcement around third-party tools using subscription OAuth tokens — tools like Agent Zero, OpenCode, and others that route requests through subscriber accounts. The reasoning is sound: fully automated agents could theoretically consume tokens around the clock, far exceeding what any subscription pricing model could have accounted for.

However, the current binary choice — subscription (human-only, no third-party agents) vs. API (full programmatic access, variable cost) — leaves a significant gap. There's a large and growing class of legitimate use cases that sit right in the middle: users who want to integrate Claude into personal agentic workflows, not to run industrial-scale automation, but to meaningfully extend how they work.

Agentic platforms are accelerating fast. Tools like Agent Zero, n8n, and others are only going to become more capable and more widely adopted. Getting ahead of this with a principled, sanctioned solution is strategically smarter — and more user-friendly — than an ongoing enforcement posture.

Proposed Solution

The Proposal

Bundle a daily rolling API token allotment with Pro and Max subscription tiers, sized using the same behavioral assumptions already baked into subscription pricing — namely that humans sleep, get distracted, go on vacation, and don't run at maximum throughput 24/7.

How it would work

One subscription, one pool, multiple access surfaces. Whether a user is chatting on claude.ai, working in Claude Code, or running an agentic workflow through a third-party tool, all consumption draws from the same underlying daily allotment. There is no separate bucket to arbitrage between.

• Each subscription tier receives a daily allotment sized proportionally to the tier, for example:
• Pro – $x equivalent per day
• Max $100 – $y equivalent per day
• Max $200 – $z equivalent per day
• Consumption through the front-end interface and consumption via API key both draw from this shared pool, in proportion
• Once exhausted within the window, the user is throttled across all surfaces — front-end and API alike — until the window resets
• Users who need more can purchase additional API tokens at standard rates, just as today

Illustrative config (conceptual)

{
  "subscription_tier": "pro",
  "daily_allotment": "$x",
  "pool": "unified",
  "surfaces": ["claude.ai", "claude_code", "api_key"],
  "on_exhaustion": "throttle_all_surfaces",
  "overage": "pay_per_token_at_standard_rates"
}

A note on reset window synchronicity

Current subscription front-end throttling operates on a roughly 5-hour window, while this proposal references a longer rolling window for the unified pool. There is an inherent asynchronicity between the two that would need to be resolved in implementation. The right mechanics — whether that means harmonizing both to a single window, or having the 5-hour front-end behavior operate as a sub-mechanism within a larger daily pool — is something Anthropic's team is better positioned to determine. The core principle here is the unified pool; the window implementation is a detail, not a blocker.

Why This Works for Anthropic

It's economically consistent — and the unified pool makes it airtight. Subscription pricing already implicitly models human behavior — sleep cycles, attention gaps, weekends. A daily allotment sized to that same model isn't a new concession; it's making the existing implicit assumption explicit and portable. Crucially, by drawing API and front-end usage from the same pool, there's no arbitrage opportunity — a user can't effectively double their consumption by using both channels simultaneously. Anthropic is simply tracking one budget per user across multiple surfaces, which is actually simpler than maintaining separate ceilings.

It closes the enforcement gap without playing whack-a-mole. Right now, motivated users are finding workarounds (OAuth token spoofing, reverse-engineered clients, etc.) and Anthropic is banning accounts reactively. A sanctioned path eliminates the incentive to circumvent, and brings those users — and their usage data — back within Anthropic's telemetry and safety monitoring.

It likely grows API revenue, not cannibalizes it. Many subscribers currently have zero API spend because the cost unpredictability of pure API access is a barrier. A small bundled allotment introduces them to API usage in a low-risk way. Users who find value will naturally top up — converting occasional API revenue from users who previously generated none.

It strengthens the third-party ecosystem. Developers building agentic platforms need a reliable, sanctioned integration path to build against. This gives them one, which in turn makes Claude more embedded across the tooling landscape — a compounding platform advantage.

Why This Works for Users

• Legitimate agentic workflows become possible without requiring a full API account or accepting unpredictable billing
• The daily reset model naturally prevents runaway costs on both sides
• Subscribers get more value from their existing plan without a price increase
• Third-party platforms can build against a supported, stable integration path

Considerations & Open Questions

• Allotment sizing — the right numbers would need modeling, but the inputs are available: average human daily token consumption, tier pricing, and acceptable margin. The key is that the allotment reflects human-pace usage, not automation-maxed usage.
• Key management — a dedicated subscription-linked API key (separate from a full Console API key) could make it easy to enforce allotment limits and distinguish this usage from standard API traffic.
• Abuse surface — the rolling window throttle and per-tier allotment sizing together make this materially harder to abuse than the current OAuth token approach, since there's a hard ceiling with no workaround short of paying for overages.

Alternative Solutions

_No response_

Priority

High - Significant impact on productivity

Feature Category

API and model interactions

Use Case Example

_No response_

Additional Context

Related Issues

This proposal is related to but distinct from #18340, which requests subscription authentication support for specific third-party IDE integrations like JetBrains. That issue addresses the authentication surface; this proposal goes further by suggesting a unified token pool model with a daily allotment tied to human usage patterns, cross-surface throttling, and a broader economic framework that makes the arrangement sustainable for Anthropic. The two issues are complementary and may be worth considering together.

Closing Thought

The agentic platform ecosystem isn't slowing down. The question isn't really whether subscribers will want to use Claude inside third-party agents — they already do, and the demand is only growing. The question is whether that happens through sanctioned, monitored, revenue-generating channels or through workarounds that create enforcement overhead and user frustration on both sides. This proposal tries to make the sanctioned path the obvious, easy choice.

Happy to discuss further or provide additional detail on any aspect of this.