Anthropic adds $20-$200 monthly Claude Agent SDK credits starting June 15

• ClaudeDevs' announcement says paid Claude plans get a separate monthly budget for programmatic usage on June 15, covering the Claude Agent SDK, claude -p, Claude Code GitHub Actions, and third-party apps built on the SDK.
• According to ClaudeDevs' plan table, that budget starts at $20 for Pro, $100 for Max 5x, $200 for Max 20x, plus per-seat credits for Team plans.
• ClaudeDevs' billing explainer adds that once the monthly credit runs out, usage can continue only through opt-in usage credits billed at API rates, or pause until the next reset.
• The loudest backlash came from wrapper and harness builders, where theo's list of affected tools and mattpocockuk's reaction both framed the change as a sharp cut for subscription-backed programmatic runs.
• Anthropic's own support page, linked in ClaudeDevs' support link, confirms the new split: interactive Claude usage keeps existing subscription limits, while SDK-backed usage moves into the new credit bucket.

You can read the official support article, skim Matt Pocock's longer breakdown, and the weirdest wrinkle is that Anthropic is simultaneously expanding support for third-party SDK apps while making many of the heaviest wrapper users feel newly rate-limited, as theo's screenshot of official comms and theo's cancellation post make painfully clear.

Anthropic's core change is simple. Anything it classifies as programmatic use stops sharing the same pool as Claude Code and chat.

According to ClaudeDevs' third-party clarification, the new bucket covers four surfaces:

• Claude Agent SDK
• claude -p
• Claude Code GitHub Actions
• Third-party apps built on the Agent SDK

The company pitched the split as a cleanup for a long-running source of confusion. In ClaudeDevs' clarification, subscription limits stay reserved for interactive use, while the support article spells out the same boundary in more detail on Anthropic's support page.

The credit schedule is tied directly to plan tier:

• Pro: $20 per month
• Max 5x: $100 per month
• Max 20x: $200 per month
• Team Standard: $20 per seat
• Team Premium: $100 per seat
• Enterprise: varies by seat type

The mechanics matter more than the marketing copy. Per ClaudeDevs' billing explainer, users claim the credit once, programmatic calls draw from it automatically, and after depletion they either switch to usage credits at API rates or stop until reset. ClaudeDevs' reset note adds that credits refresh each billing cycle and do not roll over.

Most of the anger is not about whether Anthropic finally documented SDK usage. It is about where Anthropic drew the line between interactive and programmatic.

As jeremyphoward's post puts it, the new policy effectively redefines interactive use as usage through an Anthropic front-end, not usage that feels interactive to the person at the keyboard. That matters because claude -p, headless runs, CI usage, and custom harnesses all move into the metered bucket even when a human is actively driving them.

The same complaint shows up across different users and different estimates:

• mattpocockuk called it a 10x cut to claude -p.
• mattpocockuk's longer thread said AFK Claude Code usage gets slashed by roughly 5x to 20x.
• theo's tool list said tools like T3 Code, Conductor, Zed, Jean, CI uses of claude -p, and scripts that call Claude Code from other tools all become much more constrained.
• andersonbcdefg's post said losing subscription-backed claude -p removed their preferred way to use Claude.

None of those posts changes Anthropic's stated rules. They do show exactly where the new billing boundary collides with how power users were already working.

One genuinely new thing shipped here: Anthropic is now explicitly allowing paid Claude plans to work through third-party apps built on the Agent SDK.

That means tools such as OpenClaw, Conductor, and T3 Code can draw from the monthly SDK credit instead of requiring a separate API key path, as alexalbert__ and ClaudeDevs' third-party clarification both state. For builders who wanted clean official support, that is a real policy change.

But the same section produced the harshest backlash from wrapper authors. theo's first reaction said T3 Code is now officially supported, but billed against API-style credits; charlieholtz's Conductor note told Conductor users that Max subscribers get $200 and then pay API costs; and nummanali's post called the GitHub Actions plus headless claude -p treatment a possible downgrade.

The result is a strange split-screen launch. Anthropic opened the door to third-party SDK apps, while some of the developers who built those apps immediately argued the economics got worse.

The strongest negative reactions came from people who already use Claude mostly through their own loops, automations, and UIs.

omarsar0's post says roughly 80 percent of their Claude use already runs through the Agent SDK, with custom harnesses and long-running automations instead of Anthropic's own surfaces. theo's longer post says T3 Code wrapped the Agent SDK because it was the only supported path, and that users now face a drastic reduction in usable subscription-backed volume.

That is the important technical divide in this story:

• Anthropic preserved its existing interactive subscription pool.
• Anthropic moved SDK-backed usage into a fixed monthly dollar credit.
• Developers who built their own harnesses were often treating those two things as interchangeable.

omarsar0's follow-up frames the business consequence bluntly: developers building serious agent harnesses on top of the Claude Agent SDK are starting to move work elsewhere. That does not settle whether Anthropic's policy is right. It does show which segment thinks it just got priced out.

The rollout has two fixed dates. According to ClaudeDevs' rollout post, users get an email on June 8 to claim their credits, and the new policy takes effect on June 15.

One extra operational detail sits in the claim flow itself. Per ClaudeDevs' billing explainer, usage credits that bill at API rates are optional and can be turned on or off, so the post-credit behavior depends on whether a user enables that fallback. If they do not, programmatic usage pauses until the next monthly reset.