Skip to main content
Presston AI MCP checks billing before it starts a paid AI operation. The Developer API authorizes the operation, and Boxpressd Payments reserves and settles credits from the application’s shared AI balance.

​
Non-billable tools

You can inspect billing without consuming a usage unit:
  • get_billing_status
  • get_billing_usage
  • list_billing_plans
  • list_ai_credit_packs
Ask your client:
For each public paid AI tool, the MCP server sends one idempotent request to the Developer API. The API then:
  1. Checks the api.ai entitlement.
  2. Reserves a conservative credit ceiling.
  3. Calls the protected AI service.
  4. Settles actual provider usage on success.
  5. Releases unused or failed reservations.
Actual settlement can include input tokens, cached-input tokens, output tokens, and web-search calls. Pricing policies and provider rates live in Boxpressd Payments. The MCP tool does not contain a retail price or calculate the deduction.

​
Credit-gated tools

Text recognition, fingerprint extraction, wrapper analysis, band analysis, complete identification, and cigar enrichment cross the paid operation boundary. The complete identification workflow can finish from local evidence, but its paid fallback remains credit-gated. Detection, preprocessing, local similarity search, fingerprint catalog lookup, and billing inspection do not invoke paid provider inference. They still require AI access to be enabled for the application.

​
Handle unavailable credits

When the account lacks credits or entitlement, the tool call fails before paid downstream work. Use get_billing_status or list_ai_credit_packs to retrieve the management action, then send an authorized account owner to the Developer Dashboard. Do not ask an AI model to handle card details. Credit purchases use the hosted Boxpressd Payments flow. For first-party delegated-user calls, both application and user ledgers can apply. An error can identify developer or user as the balance that blocked the operation.