How to Download OpenAI Invoice PDFs from the API Platform Automatically (2026)

Month-end closes. Your AP lead asks for last month’s openai invoice PDFs from the API platform. Someone signs into platform.openai.com, opens Billing history, and downloads PDFs one row at a time. Next month, the same hunt starts again—unless the person who had Owner access left, the org switcher was on the wrong workspace, or someone mistakenly scraped ChatGPT Plus receipts from chatgpt.com and filed the wrong product.

That last mix-up is more common than it should be. OpenAI ChatGPT billing and OpenAI API billing are separate systems. This guide is only for API / platform invoices at platform.openai.com. It is not for ChatGPT Plus, Pro, Business, or Enterprise bills on chatgpt.com. OpenAI’s own help center (article 9039756) states that clearly: manage each product’s charges in its own settings.

If you need a repeatable way to pull OpenAI API invoice PDFs into AP without writing a brittle Playwright script, this is the path.

Runs in an isolated BrowserAct browser (cloud Bot), not as a Chrome extension on your daily profile—so you can keep separate logins, proxies, and schedules.

Invoice download is an application-layer job. Tools that speed up agents on an already-logged-in daily browser and tools that give agents isolated, parallel browser infrastructure can both touch it. BrowserAct’s edge for this use case is the second model: cloud-side isolation, multi-identity, and schedulable Bots—not hitchhiking on the Chrome profile you use for email and Slack.

TL;DR / Key Takeaways

• Keep the OpenAI API Owner/Admin login in an isolated BrowserAct browser, not your daily Chrome.
• Point the Bot at Billing history on the API platform, list invoices by date range, and download original PDFs.
• Deliver files to AP (email, Drive, S3, or a shared folder) with stable filenames keyed by invoice number.
• Treat missing invoices as a first-class result: zero-usage months, wrong org, or Reader role often explain the gap.

Before you start

Have these ready before the first dry-run:

• A BrowserAct account (cloud Bot path below)
• Portal URL: https://platform.openai.com/ (API platform—not chatgpt.com)
• Billing history URL often used: https://platform.openai.com/settings/organization/billing/history
• An OpenAI API organization login with Owner or Admin billing access (Reader often cannot download)
• Your 2FA method (authenticator app, email, or SMS)
• One already-reconciled month of invoices as a ground-truth check
• Optional: AP notify email, Drive/S3 path, or shared folder for delivery
• Optional: organization name if you manage more than one API org under the same login

Do not use ChatGPT Settings → Billing for this job. That is a different billing system.

Choose your path

Keep these three lanes separate. This article’s main tutorial is Cloud Bot only.

Main how-to: Cloud Bot

1. Create or open the OpenAI API invoice Bot

In BrowserAct, create an Agent / Workflow Bot for OpenAI API invoices, or open the official template if your workspace already lists one. Keep inputs explicit so month-end runs stay boring:

• start_date / end_date (inclusive)
• organization_name (when the login can see more than one org)
• notify_email or delivery path (optional)

Paste a prompt like this into the Bot (adapt variables; keep the structure):

Open https://platform.openai.com/settings/organization/billing/history.
Sign in with the saved OpenAI API org owner/admin login in the isolated browser.
If multiple orgs, select {{organization_name}}.
List invoices/receipts with issue or billing date between {{start_date}} and {{end_date}} inclusive.
Paginate until done.
Download each original PDF (invoice preferred over bare receipt when both exist).
Name files: openai-api_{{invoice_number}}_{{issue_date}}.pdf.
Return table: invoice_number, issue_date, amount, currency, filename, status.
List any you could not download and why.
Never use ChatGPT Settings → Billing for this job.

That prompt is the whole “script.” You are teaching the Bot the portal path, not claiming OpenAI exposes a public invoice-download API for model usage. The Completions / Responses APIs are not invoice download APIs—do not wire those and expect PDFs.

2. Attach login in an isolated browser

Create a dedicated browser identity for this OpenAI API organization (or for this client, if you are an accounting firm). Bind a static proxy when the account expects a stable network.

Complete the first sign-in and 2FA yourself in that isolated browser, then save the session. This is the critical product difference: the login lives in a cloud Bot browser, not as an extension riding your everyday Chrome profile. Separate clients get separate browsers so cookies never cross.

Pro Tip: One OpenAI org (or one client) per isolated browser identity. Never reuse the same Bot browser across unrelated AP clients.

If the Bot hits a credential wall later, pause for human assist, finish MFA once, and resume—the same pattern covered in BrowserAct’s notes on login, Google Auth, 2FA, and manual takeover and workflow policies that decide when to pause.

3. Dry-run one known month

Run the Bot against a month you already closed in the books.

Check:

• Invoice count matches your manual archive
• Invoice numbers, amounts, currency, and legal entity / org name
• Open one or two PDFs and confirm they are API platform documents, not ChatGPT subscription receipts

Common fixes after the first dry-run:

• Pagination stopped early—widen “load more” / next-page waits
• Download click opened a preview tab—wait for the PDF response, then save
• Org switcher still on the wrong organization
• Role is Reader—switch to Owner/Admin or grant billing download permission
• Month had prepaid credits / auto-recharge docs only, or zero usage and no invoice row

Pro Tip: Keep a ground-truth folder for that month. If the Bot returns a different count, fix pagination/org switcher before you schedule anything.

4. Deliver files and schedule monthly

After a clean dry-run, wire delivery:

• Download artifacts from the Bot run
• Email AP, or drop into Drive / S3 / a shared folder
• Keep filenames stable: openai-api_{invoice_number}_{issue_date}.pdf
• Dedupe on invoice number so re-runs do not double-archive

Then schedule. For self-serve API customers, pull after the prior month’s charges settle. For Enterprise API customers, OpenAI notes invoices are typically issued within about two weeks after the billing cycle ends (invoice dates help article)—so schedule with that lag, not on the 1st by default. Alert on failure; keep the “could not download” list for the AP review queue.

Pro Tip: For Enterprise API accounts, build in that ~two-week invoice lag—otherwise month-end runs return empty lists and false “missing invoice” alarms.

Optional: Run from Claude or Cursor via MCP

If your team already works in Claude or Cursor, keep the same cloud Bot and call it over MCP. MCP here is not “drive my laptop Chrome.” It is a remote control plane for BrowserAct cloud Bots.

1. Connect BrowserAct MCP at https://mcp.browseract.com/mcp/
2. Ask the assistant to find the OpenAI API invoice Bot (search / list bots)
3. Supply missing inputs (start_date, end_date, organization_name)
4. Run once, then fetch task status and download the PDFs

If token burn becomes an issue while the agent chatters around the browser, keep browser state compact and prefer a packaged Bot over rediscovering the portal every time—see how to reduce MCP token usage in browser automation.

Advanced: CLI + Skills for multi-org and proxies

Use CLI + Skills when one operator manages many client OpenAI orgs, needs fixed proxies, or wants the same Skill inside a local / self-hosted agent loop.

High-level shape (details live in BrowserAct CLI docs; this is not a full install tutorial):

• Install the BrowserAct CLI and entry Skill
• Pull core Skills (for example via browser-act get-skills core)
• Run each org in its own stealth / isolated browser identity with a pinned proxy
• Reuse the same invoice prompt / Skill; swap identity + org name per client

Cloud Bot remains the default for AP month-end. CLI is the advanced lane when isolation and multi-identity are the product requirement—not a second copy of the beginner tutorial.

Why OpenAI API invoices stay manual

A few structural reasons this job still lands on a human calendar:

• PDFs live behind login; there is no public “download all invoices as ZIP” button for typical self-serve orgs
• The menu path is deep: Settings → Organization → Billing → Billing history
• Download permission is role-gated; Owner/Admin usually required
• ChatGPT and API billing look similar enough that teams pull the wrong portal
• Prepaid credits, auto-recharge receipts, and zero-usage months create uneven document sets
• Enterprise timing lags calendar month-end, so “pull on the 2nd” often returns an empty or partial list

None of those problems is solved by calling the model API harder. They are portal problems—exactly where an isolated browser Bot earns its keep.

Ways to collect OpenAI API invoices

Honest comparison. Short cells on purpose.

What this Bot returns

Result fields

Common failure reasons

• Wrong product portal (ChatGPT billing instead of API platform)
• Insufficient role (Reader / member without billing download)
• Wrong organization selected in the switcher
• Date range before invoices were issued (especially Enterprise lag)
• Zero-usage month with no invoice row
• Session expired; 2FA required again
• PDF open in a new tab / viewer; download wait too short
• Pagination incomplete