The build quote: what a model build will use, before it runs
Before a model build runs, Anticode asks the server what it's likely to cost and prints the answer under the build button: roughly what share of today's limit it will take, about how many tokens it will use, and how much of today and this week you've already used. Assistants working through the MCP server have to fetch the same quote and show it to you before they can start a build.
Why a quote comes before the build
Model builds are paid for by what the model actually reads and writes, so two requests that look similar can cost different amounts. Most AI tools only tell you afterwards, if at all, which makes it hard to decide whether a long, detailed request is worth it on a day when you've already used a lot. Anticode measures your allowance as a daily and a weekly budget, and an AI build cost estimate in front of the button turns that budget into something you can plan with: you see the likely share before you commit, not a surprise when the meter jumps.
What the line under the build button says
When you're signed in and a build can run, the note reads in three parts, separated by dots. A typical one looks like Uses about 4% of today's limit · about 2,100 tokens · 34% used today · 41% this week.
| Part | What it tells you |
|---|---|
| Share of today's limit | The estimated cost of this build as a percentage of your daily allowance, never shown as less than 1% |
| Tokens | A rounded estimate of input plus output tokens. Hover it for a note saying it's an estimate and that failed builds cost nothing |
| Used so far | How much of today's and this week's allowance is already spent |
Underneath, a second line reminds you which changes are free. Switching direction, theme, colour, sections or the site's name after it's built runs in your browser and never touches your limits. A change written in your own words goes to the model and counts like a build.
How the estimate is worked out
The builder sends only the length of your request, never the words, and caps it at 600 characters for the calculation. The server adds the fixed instructions and output schema every page build carries, the typical size of a finished content plan, and, when web research is switched on, an allowance for the search results the model reads. That total is priced and compared with your plan's daily limit.
App builds are estimated differently, because the model writes the whole program rather than a content plan. The quote assumes a mid-sized app and says so; a large one, such as a spreadsheet with formulas and charts, can use about twice as much.
When the quote says no
The same request also tells you when a build can't run right now, and why, so you find out before typing a long brief rather than after:
- Signed out. The note says you'll get a free preview laid out in your browser, and that signing in gets model-written copy. See preview without an account.
- Email not confirmed. Builds wait until you confirm your address.
- Paused after failures. Several failed builds in a row pause model builds for a few minutes to protect your allowance; the quote says so up front.
- Limit reached. It names the window you've used up, today's or this week's, and when it resets.
- Paying from credits. If your allowance is used up and you hold credit packs, it says the build will be paid from your credits and shows what's left.
Quotes for AI assistants over MCP
When Claude, ChatGPT, Cursor or another client works on your projects through the MCP server, it can't start a model build or a model change on its own say-so. It first calls get_build_quote for the project, which returns the same cost summary, whether the build can run, and a quote id. The tool's instructions tell the assistant to show you the summary and wait for your go-ahead.
- A quote id is tied to one project and one purpose: a build, or a change.
- It expires after 15 minutes and can be used once.
- If the build fails, the quote is handed back, because failed builds cost nothing.
- A read-only connection gets the cost but no quote id, so it can't start anything.
Limits of the estimate
- It's an estimate. The build is charged for the tokens it really uses, which can be more or less than quoted. The exact figures appear afterwards on the build receipt.
- Shares, not prices. The builder shows a percentage of your allowance rather than a money amount; what the allowance is on each plan is on /pricing.
- Apps vary most. A short request for a complicated app can use far more than the mid-sized assumption.
- It doesn't reserve anything. Another build started elsewhere on the account in the meantime still uses the same allowance.
Questions people ask
Does a failed or refused build use my allowance?
No. Only builds that finish successfully count. Refused and failed builds are recorded as not charged, and over MCP their quote is handed back.
Why is the token figure rounded?
It's built from typical sizes, not your actual output, so it's rounded up to the nearest hundred rather than shown with false precision.
Are quick changes quoted?
They don't need to be. Direction, theme, colour, sections and the name are changed in your browser without the model and don't use your limits.
Can an assistant skip the quote?
No. A model build or change through the MCP server needs a valid quote id for that project, fetched in the last 15 minutes and not used before.