billing — Claude Code skill: what it does, how to install
billing — a skill for Claude Code and Codex that checks Bright Data balance, inspects spend per zone, and estimates upcoming job costs.
MIT 185 installs ★ 272 updated 2026-10-15
npx skills add brightdata/skills --skill billing -a claude-codeWhat billing does
This tool reports current credit, balance figures, and historical spending on a Bright Data account using bdata budget commands and https://api.brightdata.com endpoints. It remains read-only under every circumstance: it never alters billing plans, adds funds, or purchases datasets.
Estimates are calculated beforehand without triggering billable requests, scrapers, or browser sessions. Access requires an API key carrying Finance or Admin permissions, with Finance providing the necessary least-privilege scope. Zone cost queries cover proxy networks and Web Unlocker, whereas scraper data requires /costs/export/json queries tied to dataset identifiers.
What it helps with
- check remaining account balance and pending charges
- calculate zone expenses between specific calendar dates
- project job expenses before running scraping operations
- identify which domains consumed the bandwidth budget
Example prompts for the agent
- Check the remaining budget on our Bright Data account
- What did zone
res_proxiescost from 2025-02-01 to 2025-02-15? - Estimate the cost to crawl 200 target URLs using Web Unlocker
How to install billing
1Skills CLI (Vercel)
npx skills add brightdata/skills --skill billing -a claude-code2Manually: put the files in place
~/.claude/skills/billing/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/billing/
1Skills CLI (Vercel)
npx skills add brightdata/skills --skill billing -a codex2Manually: put the files in place
.agents/skills/billing/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add brightdata/skills --skill billing -a cursor2Manually: put the files in place
.cursor/skills/billing/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install brightdata/skills skills/billing2Manually: put the files in place
.github/skills/billing/SKILL.md1Skills CLI (Vercel)
npx skills add brightdata/skills --skill billing -a gemini-cli2Platform CLI
gemini skills install https://github.com/brightdata/skills --path skills/billingManually: put the files in place
.devin/skills/billing/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add brightdata/skills --skill billing -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about billing
How to install billing in Claude Code?
Run npx skills add brightdata/skills --skill billing -a claude-code in your terminal.
What does the billing skill do?
It inspects remaining credits, retrieves spend history across zones, and computes cost estimates without modifying account settings or running live jobs.
What is required to run billing?
A Bright Data API key with Finance or Admin privileges, configured in the environment or inside the bdata CLI credential storage.
Source text
SKILL.md · MIT
--- name: billing description: 'Use when the question is about money on a Bright Data account: current balance, credit left, what was charged, spend or cost history, credits or free tier used, cost per zone, product or dataset, or what a job will cost before it runs. Read-only, and it never buys or changes anything. Not scraping or extraction (use scrape), not how to drive the CLI itself (use brightdata-cli), not a refused or failing call (use agent-onboarding).' --- # Bright Data - Billing Two questions come up: what has this account already spent, and what will this job cost? The first has an exact answer from the account's own numbers. The second is always an estimate, so say so. ## Read-only, no exceptions Read the numbers, report them, stop. Never add or change a payment method, never top up a balance, never set auto-recharge, never change a plan, never buy a dataset, never cancel anything. Those actions belong to the person, not the agent, even when the user asks the agent to do them. Send them to brightdata.com/cp/billing/settings and let them click it themselves. Never run a real job to find out what the account can do or what something costs. A Web Unlocker request, a scraper trigger and a browser session all spend money. Every read below is free and starts nothing. ## The reads | The question | Command | Endpoint behind it | |---|---|---| | How much money is left | `bdata budget balance` | `GET /customer/balance` | | What one zone cost | `bdata budget zone <name> --from <date> --to <date>` | `GET /zone?zone=<name>` and `GET /zone/cost?zone=<name>` | | What all active zones cost | `bdata budget zones --from <date> --to <date>` | `GET /zone/get_active_zones`, then one `/zone/cost` per zone | | What each product, zone, dataset or domain cost | no CLI command, call it directly | `POST /costs/export/json` | | What scraping cost | no CLI command, call it directly | `POST /costs/export/json` with `dimension` `web_apis` for Web Scraper API, `collectors` for Scraper Studio | | Which zones exist, including deleted ones | no CLI command, call it directly | `GET /zone/get_all_zones` | | Which website ate the budget | no CLI command, call it directly | `GET /domains/req` and `GET /domains/bw`, dates required | | How many records a snapshot delivered | no CLI command, call it directly | `GET /datasets/v3/snapshots?dataset_id=<id>`, read `dataset_size` | Base host is `https://api.brightdata.com` and auth is the header `Authorization: Bearer <key>`. Read the key from the environment or the CLI's own store. The exact file path for each OS, and the safe way to read it without printing it, live in the `agent-onboarding` skill's auth reference. That reference also has the thing that trips agents here: after a CLI login, `BRIGHTDATA_API_KEY` is unset, and that is normal rather than a sign the machine is logged out, because login writes the key to the CLI's store and never to the environment. Never ask the user to paste a key into the conversation, and never print one: to check that a key exists, run `node scripts/check-auth.mjs` from the `agent-onboarding` skill, never `cat`, `echo` or `jq` the credentials file or the variable. `/zone/cost` only sees zone-backed products such as Web Unlocker, SERP, Browser API and the proxy networks. Web Scraper API and Scraper Studio spend is keyed by dataset id and collector rather than by zone name, so it never appears there. Those go to the cost export. ## Which key can read this API keys carry one of five permission levels: Admin, Finance, Ops, User and Limit. Billing and cost data needs Finance or Admin, and Finance is the least-privileged one that works, so prefer it. Use a Finance key directly in the `Authorization` header for these reads: `bdata login --api-key` expects a full-access key, so keep the CLI logged in with an Admin key and the Finance key for the API. Only account admins create or change keys, at brightdata.com/cp/setting/users. A 401 means the key is missing, revoked or wrong. A 403 means the key is real but under-privileged for billing, so name the missing access instead of retrying. A Finance key may not read product data such as snapshots or Studio jobs, and that split is fine: report the half you can read and say which access the other half needs. A 404 means a wrong or expired snapshot, job or dataset id, while a wrong zone name answers 422. ## Using the budget commands well Three habits make the `bdata budget` output reliable. Always pass `--from` and `--to` to `budget zone` and `budget zones`, so the figure covers exactly the window you asked about. For the pending charge, read `pending_costs` from `GET /customer/balance` directly. And when you need machine-readable output, call the endpoint rather than the balance command. Before writing any parsing code, read `references/cost-api.md` for the field lists and the live response samples, including how the date parameters behave (`to` is exclusive on these endpoints). ## What a job will cost before it runs No endpoint returns a quote, so never present one as if it came from the API. Build the estimate in three steps and label it an estimate. 1. **Find the billing unit for the product.** Web Unlocker and SERP bill per successful request, and failed requests are not charged, with one exception: enabling custom headers or cookies bills every request. A SERP request that returns several result pages is still one request. Web Scraper API bills per delivered record, and attempts that fail because of incorrect inputs still bill. Scraper Studio bills per page load, with file downloads charged separately per GB. Browser API and the proxy networks bill per GB of traffic, and some proxy plans are a fixed charge per IP instead. 2. **Count the units, not the inputs.** One input is not one record. Five listing URLs that each hold twenty products produce a hundred records. A batch of ten URLs is ten separate billable requests. Pagination pages, detail pages and extra navigations each count as page loads in Scraper Studio. 3. **Use the account's own rate, not the public price list.** Accounts on custom pay-as-you-go or pre-commit plans have their own rates, so the public list may not apply. For Web Unlocker and SERP you can compute the real rate: ask `/zone/cost` for a period with `from` and `to`, then divide the `custom` bucket's `cost` by its request count. When the bucket carries both `reqs_unblocker` and `reqs_serp`, divide by their sum, because the cost is one combined figure for the zone (live). For Web Scraper API the record count is a read (`GET /datasets/v3/snapshots?dataset_id=<id>`, `dataset_size`) and the dollars are in the cost export, per dataset under `web_apis` (live) or per snapshot under `ws_api_snaps` where attributed, so compute the real per-record rate when the reads return data. If the account has no past usage, read its own plan from `GET /pricing_plans` (Admin or Finance key; the plan with `is_current: true`, its `product_prices`, where `req` is per 1,000 requests). Only if that read is refused, open the public list at `brightdata.com/pricing` (for Web Scraper API, `brightdata.com/pricing/web-scraper`) and say it is the list price, not this account's rate. Never quote a price from memory, since prices change. In the reply, name the source you read. ## Free credits Every eligible account gets 5,000 free credits a month from one shared pool. They reset on the first of the month and unused credits are forfeited rather than carried over. Web Unlocker, SERP and Web Scraper API spend one credit per request or record, and Scraper Studio spends one per page load. The Browser API draws on the same pool at 5 credits per MB of traffic (in effect from 2026-09-01), so the full monthly allowance covers close to 1 GB of browser traffic. Accounts on custom pay-as-you-go plans and on pre-commit plans get no monthly free credits. The proxy networks are not covered. New accounts instead get a one-time $2 trial credit for the proxy products, valid 7 days, and a further $5 for adding a payment method, valid 30 days. All of these figures are from the docs. The registration completion call returns a brand-new account's `entitlements` (`monthly_credits`, `trial_credit_usd`, `trial_days`) once, at signup (the `agent-onboarding` skill, No account yet). None of the billing reads above return them. **Free credits are not in any billing API.** No balance field counts them, and the `credit` field on `/customer/balance` is dollars, not a credit count. Free-tier and trial state live in the control panel at brightdata.com/cp/billing/overview under Free Tier Credits, which also shows the renewal date. Send the user there rather than guessing, and never read a money balance of zero as proof that the free credits are gone. They are separate pools. ## What an MCP tool call bills as For "what did my agent's tool call cost". Which tools are turned on (`groups=`, `tools=`) never changes what anything costs. `search_engine` and its batch variant bill as one request per query, `scrape_as_markdown`, `scrape_as_html` and `scrape_batch` as one Web Unlocker request per URL, `run_scraper` as Web Scraper API records delivered, and `scraping_browser_*` as Browser API bandwidth. The credential makes no difference: a Remote client on `?token=` and one on OAuth 2.1 reach the same tools and draw on the same account credits, so an OAuth client spends the credits of whichever Bright Data account its user signed in to, which may not be this one. ## Read next - **Before calling any of these endpoints:** [references/cost-api.md](references/cost-api.md) - exact paths, parameters, real response shapes, the eleven cost dimensions, and which facts are verified live against an account rather than only read in the docs. - **When a read is refused:** the `agent-onboarding` skill. A missing or invalid key gives 401 with a short plain-text body on these endpoints, and `No API key found` from the CLI. Both mean log in. ## Red flags - Adding a payment method, topping up, or changing a plan on the user's behalf - Running a real job, or a single test request, to find out what something costs - Reading `pending_balance` from a live balance response instead of `pending_costs` - Treating every key of a cost export as a date, so `total` gets added to the days - Reporting a zone cost without `--from` and `--to` - Adding `back_m0` to `back_d0`, or assuming `bw` is present in every bucket - Reporting an empty cost export as proof of no spend, without the `products` cross-check described in `references/cost-api.md` - Quoting public list prices as though they were this account's rate - Asking the user which plan or zone they use when `/zone/get_all_zones` and `/zone?zone=` answer it - Presenting an estimate as if it were a quote from the API, which has no estimate endpoint - Echoing a raw `/zone?zone=` response, which contains the zone password. `bdata budget zone <name>` prints only a summary without it (on CLI 0.4.1 its `--json` and `--pretty` are taken by the parent `budget` command and ignored); use it, or `/zone/cost`, for cost - Printing an API key, or asking the user to paste one into the chat
© brightdata · MIT · brightdata/skills