Table of Contents
Claude Agent SDK credit limit (June 2026): what n8n, Make, and Zapier users need to do before June 15
Starting June 15, 2026, Anthropic stops counting Claude Agent SDK and claude -p usage against your subscription's normal limits and routes it to a separate monthly credit instead. If you run automations that call Claude, the first thing to do is check how those automations authenticate, because the answer decides whether this change touches you at all. The short version, after reading Anthropic's own support page and the Claude connection docs for all three platforms: the standard Claude nodes in n8n, Make, and Zapier connect with a direct Anthropic API key, and API-key usage is explicitly out of scope. The credit pool only bites if you run the Agent SDK or claude -p as a custom step inside your workflow. This guide shows you how to tell the difference, what the credit buys, and the one decision worth making before the deadline.
The two billing pools, explained
Anthropic is splitting Claude subscription usage into two buckets. One bucket is interactive use: chatting with Claude on the web, desktop, or mobile, and running Claude Code live in your terminal or IDE. That bucket keeps drawing from your plan's usage limits exactly as it does today. Nothing there changes.
The second bucket is automated, non-interactive use. Per the Anthropic support page, that means the Claude Agent SDK in your own Python or TypeScript projects, the claude -p command (Claude Code's non-interactive mode), the Claude Code GitHub Actions integration, and third-party apps that authenticate with your Claude subscription through the Agent SDK. After June 15, this bucket no longer draws from your subscription limits. It draws from a new monthly Agent SDK credit, which you claim once and which refreshes every billing cycle. When that credit runs dry, requests either continue at standard API rates (if you have usage credits enabled) or stop until the credit resets.
One line on that page settles the question for most automation users: "If you use the Agent SDK with an API key from the Claude Platform, nothing changes." Direct API-key billing is pay-as-you-go and always has been. It never touched your subscription pool, so the split does not move it.
Credit amounts by plan
Here is the full table, pulled directly from Anthropic's support page. The last column translates each credit into rough Claude Sonnet token volume so the dollar figure means something. Sonnet 4.5 runs $3 per million input tokens and $15 per million output tokens per the Claude pricing docs, so a typical agent turn that reads a few thousand tokens of context and writes a few hundred back lands in the low cents. The "approx. runs" column assumes a modest agent turn of about 8,000 input plus 1,500 output tokens, roughly $0.045 each.
| Plan | Monthly Agent SDK credit (source) | Approx. Sonnet agent turns at API rates |
|---|---|---|
| Pro | $20 | ~440 |
| Max 5x | $100 | ~2,200 |
| Max 20x | $200 | ~4,400 |
| Team (Standard seats) | $20 | ~440 |
| Team (Premium seats) | $100 | ~2,200 |
| Enterprise (usage-based) | $20 | ~440 |
| Enterprise (seat-based Premium seats) | $200 | ~4,400 |
Example only. The per-turn cost is illustrative. A turn that loads a 40,000-token document and writes a long summary costs far more, so a heavy workflow burns the credit much faster than this table implies. Use it to size the order of magnitude, not to budget to the dollar.
Two more details from the same support page catch people out. Members on Standard seats of a seat-based Enterprise plan are not eligible to claim the credit at all. And credits are per user, not pooled. Your teammate's unused $20 cannot cover your overflow, and unused credit does not roll over to next month.
What this means for your automation platform
Before you touch anything, run the one check that decides your exposure: open the Claude connection inside your automation tool and look at how it authenticates. If it asks for an API key (the long string starting with sk-ant-), you are on the direct-API path and the June 15 change does not affect that node. If it instead logs you in through your Claude subscription, you are on the Agent SDK path and the credit pool applies.
n8n
n8n's Claude integration is the direct-API kind. The n8n Anthropic credentials docs list exactly one supported authentication method: API key. The Anthropic node and the Anthropic Chat Model node both take a key you generate in the Claude console, which means your existing n8n workflows bill pay-as-you-go and sit outside the credit split entirely.
Where June 15 reaches n8n users is the Execute Command node. Shell out to claude -p "..." from inside a workflow on a machine logged into your Claude subscription, and that call now lives in the Agent SDK bucket. Self-hosters running scheduled claude -p jobs are the group to watch. Want to keep your normal subscription limits for interactive Claude Code while metering the automated calls? Claim the credit and set an alert. Happy with pay-as-you-go instead? Switch that step to a plain API call through the Anthropic node and skip the credit pool. You can start a free n8n account and use the Anthropic node with a console key to stay on the predictable path.
Make
Make is the same story. The Make Anthropic Claude docs walk you through obtaining an API key in your Claude console and pasting it into the connection's API Key field. That is a direct Anthropic API connection, so every Make scenario calling Claude today bills pay-as-you-go and is untouched by the split. Make even ships a Simple Text Prompt module that runs without any Anthropic account, billed through Make.
So Make users have nothing to claim and nothing to migrate for their standard Claude modules. Your one action item is to confirm you are not also running a claude -p or Agent SDK step on a subscription-authenticated machine somewhere upstream of Make. Got your whole Claude path running through the Make connector with a console key? Then you are done. To wire up a fresh scenario on the same direct-key footing, create a Make account and connect Anthropic Claude with a console API key.
Zapier
Zapier rounds out the set. Its integration FAQ states plainly that "you'll need access to the Anthropic API and an API key to use Claude in Zapier," and you pick the model (Sonnet, Opus, Haiku) inside the Zap. Direct API key, pay-as-you-go, out of scope. A Zapier user whose only Claude touchpoint is the Anthropic action has no June 15 work to do. If you are building a new Claude-powered Zap, set up Zapier and connect Anthropic with a console key the same way.
The honest read across all three: the "n8n/Make/Zapier credit crisis" framing does not hold, because none of their first-party Claude connectors use subscription auth. The change is real and it is sharp, but it lands on developers running the Agent SDK or claude -p directly, not on no-code automation users wiring in a console key.
Three options if you are actually in scope
If your check came back the other way (a workflow step authenticates through your Claude subscription rather than an API key), here are the three moves, in order of least to most effort.
Option 1: Claim the credit and live inside it. Best for low-volume automation. If you fire fewer than a few hundred Agent SDK turns a month, the Pro $20 or Max $100/$200 credit (per the plan table on Anthropic's support page) likely covers you with room to spare. Claim it once, set a usage alert, and forget it. The credit drains first before any other source, so you are not accidentally spending elsewhere until it is gone.
Option 2: Enable usage credits for overflow. Best for high-volume work where you would rather not babysit a cap. With usage credits enabled, Agent SDK requests keep running at standard API rates once the monthly credit is exhausted, so nothing breaks mid-run. The trade is that you are now paying full API rates past the credit line, which is exactly the cost change Zed flagged when it noted subscriptions had been subsidizing agent usage at roughly 15 to 30 times API pricing. Without usage credits enabled, requests simply stop until the next cycle, which is the failure mode to avoid in production.
Option 3: Move the step to a direct API key. Best for any production automation. Generate a key in the Claude console, point your claude -p call or Agent SDK client at it instead of subscription auth, and the credit pool stops applying. You pay API rates from the first token, but you get pay-as-you-go predictability, no monthly cap to claim or watch, and the same billing model n8n, Make, and Zapier already use for their connectors. Anthropic's own guidance for teams running shared production automation says to use a Claude Platform API key for "predictable pay-as-you-go billing." For anything that has to run reliably, this is the cleaner setup.
A direct API key call looks like this from inside an Execute Command or shell step. No subscription auth, so it never touches the credit pool:
export ANTHROPIC_API_KEY="<YOUR_CONSOLE_KEY>"
curl https://api.anthropic.com/v1/messages \
-H "x-api-key: $ANTHROPIC_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "content-type: application/json" \
-d '{
"model": "claude-sonnet-4-5",
"max_tokens": 512,
"messages": [{"role": "user", "content": "Summarize this support ticket: <TICKET_TEXT>"}]
}'Swap that in for a subscription-authenticated claude -p step and the June 15 split is a non-event for that workflow.
Before June 15 checklist
- Open the Claude connection in each n8n/Make/Zapier workflow and confirm whether it uses an API key (out of scope) or subscription login (in scope).
- If any in-scope step matters to you, claim your Agent SDK credit once from your Claude account, or move that step to a direct console API key.
- Set a usage alert so an exhausted credit does not silently stop a production run.
How to claim your Agent SDK credit
If you decide to keep an in-scope workflow on subscription auth, the credit is a one-time opt-in. Per Anthropic's support page, eligible Pro, Max, Team, and Enterprise users claim the credit through their Claude account once, and after that it refreshes automatically with each billing cycle. Team and Enterprise admins do not claim on behalf of the org; each eligible user claims their own, and Anthropic says eligible users will get an email with claim instructions before June 15. Standard-seat members of a seat-based Enterprise plan are not eligible and will not see the option. There is no pooling, no transfer, and no rollover, so claim it on the account that actually runs the automation.
If you run fewer than a few dozen Agent SDK or claude -p calls a month, the free credit almost certainly covers you. Claim it, set a usage alert, and move on. If you run production automations on any schedule that matters, skip the credit-pool bookkeeping and switch those steps to a direct Anthropic API key now: you get pay-as-you-go billing with no monthly cap to track, which is the same footing your n8n, Make, and Zapier connectors already run on. And if your only Claude touchpoint is one of those three connectors with a console key, you are already done. The June 15 change does not reach you, and there is nothing to claim.