Claude's Linked Billing Lets Customers Pay Directly. Should Your Bot Use It?
Anthropic's Linked billing beta, launched November 17, 2026, lets a third-party app's customers pay for Claude API usage from their own Claude organization instead of sharing an API key. For a small-team bot that calls Claude on behalf of customers, this removes the token-reselling problem, but only for customers who already have a Claude org billed directly by Anthropic, and only outside a short list of excluded request types.
AI-assisted editorial: researched and drafted with AI using the primary sources linked below. Check the cited evidence and any stated test scope. Illustrative examples are not measured product results. Editorial policy.
Key takeaways
- Linked billing only works for customers whose Claude organization pays Anthropic directly; cloud-marketplace billing and subscription-only plans get a 403, so check eligibility before you build the flow.
- Failures are mostly silent: a customer can unlink with no webhook or email to you, and a wrong profile ID bills the wrong organization with no error.
- Message Batches and billable code execution can't be billed to a customer at all, so those calls must stay on your own account even after a customer links.
The decision in one line
Use linked billing if your bot's customers are other businesses that already run their own Claude Platform organization paying Anthropic directly, and you want to stop estimating, marking up, and reselling their token usage. Skip it if your customers are individual consumers, pay for Claude through a cloud marketplace, or hold only a Claude subscription plan, because none of those can be billed this way.
The mechanism is an OAuth connection, not a different API. Anthropic describes it plainly: Linked billing is now in beta on the Claude Platform. Third-party platforms can let customers pay for Claude API usage without sharing an API key. That single sentence is also the whole pitch: no key changes hands, and the customer's existing Anthropic billing relationship absorbs the usage your bot generates for them.
The catch that decides whether this applies to you at all is eligibility. Linked billing works for customers whose Claude organization pays Anthropic directly for its API usage. It doesn't work for organizations billed through a cloud marketplace. If most of your small-team customers are consumers or companies without their own Anthropic account, this feature has nothing for you yet.
Sources: Linked billing on the Claude Platform | Claude by Anthropic, Claude Platform OAuth - Claude Platform Docs.
How the link actually works (illustrative example)
Illustrative example: imagine a small-team bot called "Fieldstone" that triages support tickets using Claude on behalf of several client companies. Today Fieldstone holds one Anthropic API key and bills each client a flat monthly fee that bakes in estimated token cost. With linked billing, each client could instead link its own Claude organization to Fieldstone, and Fieldstone would stop carrying their token spend on its own account. This is a hypothetical walkthrough of the documented flow, not a measured BotBento result.
Setup happens once, in the Claude Console. A member of your Claude organization with the Admin role creates an OAuth app in the Claude Console and copies its client ID. The app holds the name that people see on the approval page, and the address in your product that they return to. Instead of a client secret, authentication runs through workload identity: your back end proves that a request comes from your app with a short-lived token from your own identity provider, such as AWS, Google Cloud, Microsoft Entra ID, or Kubernetes.
Once a client admin approves the connection, Fieldstone decides who pays on every single request. You choose on each Claude API request whether the customer pays or you pay. If a request you mark for the customer can't be billed to them, the Claude API rejects it rather than billing you. That per-request switch, the billing_policy field with bill_to set to user_profile, is the entire integration surface beyond the one-time OAuth setup.
The client keeps full visibility and control on their end without touching Fieldstone's code. This article explains linked billing, which lets an administrator (Admin or Billing member) link your Claude Console organization to an app. You pay Anthropic directly for your usage in that app, and you never share an API key. They set a monthly spend cap and can unlink from their own Console page at any time.
Sources: Claude Platform OAuth - Claude Platform Docs, Use your Claude Platform account in a third-party app | Claude Help Center.
What breaks in production
Three failure patterns matter more than the happy path. First, unlinking gives you no notice. The customer-facing help article is explicit that control sits entirely with the customer: One approval covers your whole organization. The link belongs to the organization, so it stays active if the approver leaves. The reverse is also true in practice: when a customer unlinks, your next request billed to them simply fails. Anthropic's reference documentation describes the resulting state plainly: after a connection ends, requests that try to bill that customer get refused, and the profile shows as unlinked.
Second, a misdirected request bills the wrong organization with no error raised. The operator credential behind your app can bill any customer your app has linked, so a bug in the profile ID you send is invisible until someone notices an unexpected charge. The documented fix is to scope access per connection so a given credential can only ever bill its own customer, rather than relying on your operator credential for every request.
Third, two categories of calls cannot be billed to a customer under any circumstance. A request that bills your customer (bill_to: "user_profile") can't use billable code execution features (the API returns a 400), so send those requests with bill_to: "self" or without billing_policy. Batched requests are excluded entirely: Message Batches (/v1/messages/batches) doesn't currently support billing your customer. Send the requests in a batch with bill_to: "self" or without billing_policy, and they're billed to you as usual. A request in a batch that sets bill_to: "user_profile" is refused: its result is errored with an invalid_request_error, neither you nor your customer is billed for it, and the batch's other requests run as usual. For a bot that does any batch summarization or code execution work for customers, that cost stays on your account regardless of linking.
The error codes themselves are intentionally vague. The API reference lists error code linked_organization_billing_unavailableHTTP status 402 · Anthropic cannot currently bill the organization linked to the request's user profile, for a reason that lies with that organization's own account. The response does not say what the reason is. and a separate spend-limit code with the same lack of detail. Your bot needs to treat any 402 or 403 on a customer-billed request as "stop and ask the customer," not as a transient error worth retrying blindly.
Sources: Use your Claude Platform account in a third-party app | Claude Help Center, Linked billing - Claude Platform Docs, Create a Message - Claude API Reference.
When to skip it
Skip linked billing if your bot's value depends on data-handling guarantees that need to follow the customer, not the app. The eligibility and scope limits are explicit about where the request actually runs. The OAuth overview is direct about the eligibility boundary that rules out most consumer-facing bots: organizations paying through a cloud marketplace are excluded outright, and so are organizations holding only a subscription plan rather than direct API billing.
Also skip it, for now, if your bot handles regulated data. Anthropic's own marketplace listing guidance for partner products warns against linking in that case: don't connect customers who need HIPAA, FERPA, or customer-managed encryption key coverage for the usage in question, because those protections belong to the customer's organization and do not travel with a request that runs inside your app's organization.
Finally, treat the beta label as real risk, not boilerplate. The OAuth overview states its production status plainly: it is prerelease software whose endpoints, parameters, and behavior can change, and it may not be suitable for production use. For a small team with one or two engineers, that's a reason to pilot linked billing with a single low-volume test customer before wiring it into your main billing path.
A checklist before you flip the switch
Before connecting a customer, confirm these four things against your own customer base and your bot's call patterns. Getting any of them wrong either blocks the connection outright or silently leaves cost on your own account.
- Does the customer's Claude organization pay Anthropic directly, with prepaid credit or an invoice, rather than through AWS, Google Cloud, Azure, or a subscription plan alone?
- Does any of your bot's work for that customer use Message Batches or billable code execution? If so, that portion stays on your bill regardless of linking.
- Does the customer require HIPAA, FERPA, zero data retention, or customer-managed encryption keys for this usage? If yes, the request running inside your organization won't carry those terms.
- Do you scope billing credentials per connection, so a bug or a copy-paste error can't bill the wrong customer with no error returned?
Primary sources
Sources checked 2026-11-22. Standards and product documentation can change; follow the linked version when implementing.
- Linked billing - Claude Platform Docs — Anthropic
- Claude Platform OAuth - Claude Platform Docs — Anthropic
- Create a Message - Claude API Reference — Anthropic
- Linked billing on the Claude Platform | Claude by Anthropic — Anthropic
- Use your Claude Platform account in a third-party app | Claude Help Center — Anthropic
BotBento is in development. Suggest a correction.