13-Week Cash Flow Forecast

Answers the question people actually open a cash flow tool to ask: which week does the cash go negative, and by how much?

Six figures in — opening cash balance, expected receipts, supplier payments, payroll, loan instalments, and VAT or taxes due. Out comes the 13-week grid, the break-even week, the peak funding need across the period, the levers that would move the date, and a live Excel workbook with real formulas you can keep working in.

Don't know your receipts? Give revenue and DSO instead and it converts. Same for purchases and DPO on the payables side. One number per line if your weeks are flat, thirteen if you have the detail.

Three ways to use it

1. Desktop extension — nothing leaves your machine

The best version, and the one to pick if the figures are a client's. The server runs locally and writes the workbook straight to your disk. No account, no token, no upload.

Download cashflow13-*.mcpb from Releases, then in Claude Desktop: Settings → Extensions → drag the file in.

2. Remote server — the shortest path

No account and no token. In Claude Desktop, Settings → Connectors → Add custom connector:

https://mcp.chiriba.it/mcp

In Claude Code:

claude mcp add --transport http cashflow13 https://mcp.chiriba.it/mcp

Listed in the official MCP registry as it.chiriba/cashflow13.

3. Apify Actor

Apify Standby requires your own Apify API token — that is the platform's rule, not ours. Append it to the URL:

https://ledgerworks--cashflow13.apify.actor/mcp?token=YOUR_APIFY_TOKEN

If you would rather not have an account at all, use one of the two above.

What it gives back

• the week-by-week grid with the closing cash balance
• the first week the balance turns negative, and the shortfall in that week
• the peak funding need over the 13 weeks — the cumulative low point, which is a different number from the first shortfall, and the one your bank will ask for
• levers, each with the weeks it buys and the change in funding need: collecting 15 days earlier, paying suppliers 15 days later, moving payroll a week

The levers never make an outflow disappear. Moving payroll a week pushes the last payment past week 13, and the answer says so — deferred, not saved.

The workbook

Yellow cells are inputs. Everything else is a formula. Change a weekly figure and the grid, the break-even week and the funding need all move; change the start date in one cell and all thirteen weeks follow. It recalculates on open, because no cached values are written — it is the model, not a picture of one.

Weeks end on Friday by default: US practice, and where payroll and bank cut-offs fall. Configurable, and written into the workbook header so whoever receives the file knows what convention they are reading.

Your figures

No database, no history, no account, nothing kept.

• Desktop extension — the numbers never leave your computer.
• Remote server — the workbook is held in memory for 60 minutes, never written to disk, then forgotten.
• Apify — the workbook sits on Apify's storage for at most 20 minutes, then is deleted whether it was downloaded or not.

Development

npm install
npm test          # tests are written before the code
npm run typecheck
npm run build
npm run bundle    # builds the .mcpb desktop extension
npm run release   # full release chain; stops before anything leaves the machine

Rules that shaped this

• No user-facing string is hardcoded — it goes in the localization table, in English and Italian. A test fails if a key is missing a language, or if an Italian string is a copy-paste of the English.
• Tests before code. For a financial tool, correctness is the product. A balance of exactly zero is not negative. Money rounds to the cent at every step, because 0.1 + 0.2 does not make 0.3 and a plan that says "you are overdrawn" when you are not is worse than no plan.
• No storage. That is also the GDPR answer.
• One engine, many tools. The next tools are new functions on this server, not new projects.

Licence

Business Source License 1.1. In plain terms:

• Use it for anything, including your own professional work — an accountant or controller running this for a client, and billing that client, is fine.
• You may not hand it to third parties for a fee so they can use it themselves — not as a product, not as a hosted service, not embedded in other software.
• Four years after each version is published, that version becomes Apache 2.0 and the restriction lapses.

Not OSI-approved, deliberately: the tool is free to use and stays free to use, but the option to charge for it later is kept open. For other arrangements: info@chiriba.com.

Also in this repository

Dataset Audit — you paid per result; it tells you what you got. Duplicates, records that ignored the filters you asked for, and the real cost per usable record, with a live Excel receipt. Free. Source in actors/dataset-audit/, same engine as this one.