Anthropic launched linked billing on the Claude Platform on November 17, in beta for now. It lets third-party tools like coding agents or automation platforms have their customers pay for Claude usage without an API key changing hands.
How it works
Until now, it usually went like this: you paste your API key into a vendor’s app, and the app sends its requests with your key. Anyone holding that key can run up your bill, so the vendor has to guard it like a password. With linked billing, you link your Claude Platform organization to the app instead. A Console user with the Owner, Admin or Billing role approves the link once for the whole organization. Usage goes onto your existing Anthropic billing, either drawing down a commitment or pay-as-you-go. It only works if your organization pays Anthropic directly for API usage.
You can set a monthly spend limit for each platform, or a default that applies to all of them. The Console shows what each linked app is using, and you can unlink at any time.
What platforms get out of it
Vendors no longer have to store, encrypt or rotate customer keys. Anthropic hosts the approval flow and the billing. According to the release notes, platforms set the billing_policy parameter on each /v1/messages or /v1/chat/completions request to decide who pays, using the user-profiles-billing-2026-10-01 beta header. The Python (2.10.0) and TypeScript (1.10.0) SDKs picked up the parameter the same day, along with new error codes such as one for a reached spend limit.
Nineteen platforms support it at launch, including Asana, ClickHouse, Linear, Mintlify, Railway, Ramp, Tines and Warp.
The catch is in the fine print
Requests run inside the platform’s organization. That means your own organization’s settings don’t apply there: IP allowlists, your own encryption keys and data retention rules are left out. For solo developers that hardly matters. If your company runs Claude under strict requirements, linked billing only connects the bill. That’s worth knowing before an admin approves the first app.