One double-entry ledger derived from the books you already keep, proved to zero to the minor unit, with every line naming the document it came from. It reads six books this collection already keeps, all read-only: the invoice ledger, the credit notes and purchase orders, the deposits, the expense ledger, the bank import and the fixed asset register. For a period in one currency it derives a debit and a credit for every movement in them. Revenue and VAT output come off the invoices, receivables and the payments that clear them, deposits held as the liability they actually are, expenses by category with the VAT taken OUT of the gross rather than added on top of it, fixed assets as they enter service, and depreciation charged by month. The bank import posts nothing at all, which is the decision the whole ledger rests on: a bank line and a payment record are not two transactions, they are one transaction seen twice, so cash is posted from the documents, which are the only rows that carry a second leg, and each bank row is matched to a posted cash movement of the same amount, the same direction and a date within three days, written onto the line as evidence. What is left unmatched is the output that matters, because a bank debit with no expense behind it is a payment nobody entered. Nothing is ever balanced with a plug: an entry whose own legs do not add up is posted exactly as the document states it and the difference is raised by name, with the entry, the source server and the source document behind every unit of it, because a trial balance can only find a broken document if it is allowed to come out non-zero. Currencies are never added together, a purchase order stays a memo and is never posted, a deposit applied to an invoice moves the liability rather than cash, and the ledger opens at nothing, since an opening figure nobody can walk back to a document is the first invented number in a set of books. Every line carries the server, the document id and the ISO date it came from, and the bytes and mtimes of all six sibling stores are unchanged across all six tools.
In agent mode ledger_build and trial_balance are reads over six stores and change nothing outside this server's own register, so an agent can build a period, find the imbalance, open the source document the offender names and re-derive in one turn. Every line carries its source server and document id precisely so that loop terminates somewhere real.
| You say | Tool |
|---|---|
| Build the double-entry ledger for June and tell me whether the trial balance comes to zero. | ledger_build |
| Show me every ledger line that hits receivables, with the invoice each one came from. | ledger_lines |
| Close June: what does the month leave unposted or inconsistent? | month_close |
Posting the bank import alongside the documents, which is the obvious way to build a cash book and the way most spreadsheets do it, double-counts almost all of the cash and the trial balance still comes to zero. Measured on this server's worked month: four of the five bank rows are the same money as a document already posted, 1,375,300 of 1,380,300 minor units of cash movement, 99.6 percent. Posting them as well moves the cash balance from -10,543.00 to -21,111.00 EUR, and nothing flags it, because every duplicated receipt arrives with its own contra, every individual line is plausible and the only symptom is one account off by nearly a factor of two in a direction nobody audits. What is left after matching is 7,500 minor units, one unexplained withdrawal. That 0.4 percent is the entire reason to import a statement at all, and it is exactly what a duplicate-heavy ledger buries.
The file is .vscode/mcp.json, and the key inside it is servers.
| Scope | Path |
|---|---|
| This workspace | <workspace>/.vscode/mcp.json |
| Every workspace | the user profile mcp.json, opened by the MCP: Open User Configuration command |
// .vscode/mcp.json (the key is "servers")
{
"servers": {
"cash-book": {
"command": "npx",
"args": ["-y", "@theluckystrike/mcp-cash-book"]
}
}
}Or from a terminal:
code --add-mcp '{"name":"cash-book","command":"npx","args":["-y","@theluckystrike/mcp-cash-book"]}'VS Code asks you to confirm you trust the server and its capabilities before it starts. Nothing runs until you answer.
The npm publish of @theluckystrike/mcp-cash-book is pending; until then use the .mcpb bundle or a clone and build from the latest release.
The same server runs at https://mcp.zovo.one/mcp/https://mcp.zovo.one/mcp/cash-book over MCP streamable HTTP, no install. Mint a free token with curl https://mcp.zovo.one/mcp/token, or use a Pro key. It has no filesystem, so a file comes back as a one-hour download link.
{
"servers": {
"cash-book": {
"url": "https://mcp.zovo.one/mcp/https://mcp.zovo.one/mcp/cash-book",
"headers": { "Authorization": "Bearer <token>" }
}
}
}The top-level key is "servers", not "mcpServers". A config pasted from a Claude Desktop or Cursor README parses cleanly, contributes no servers, and gives you no error to read, which makes it the most expensive one-word mistake here. Alongside it an "inputs" array holds secrets. The newer Agent Host reads a workspace .mcp.json instead.
Free: trial_balance and ledger_lines are free and unlimited on every tier. Whether the books add up is the question this server exists for, and a free tier that hides the answer is a demo rather than a tool, so neither the check nor the evidence behind it is ever metered. The cap is on the PERIOD instead: three distinct periods a calendar month, keyed by from, to and currency, so rebuilding a period already in the register is free forever on every tier and the meter only ever moves when you ask a genuinely new question. Pro is $19 once, verified offline.
The key is servers, not mcpServers. A config copied from a Claude Desktop README parses cleanly and contributes nothing. Rename it. On the newer Agent Host, use a workspace .mcp.json instead.
No. code --add-mcp takes the server object on the command line, and MCP: Add Server in the palette asks whether the target is Workspace or Global.
In Chat, behind the tools picker, once the server has started. Adding or changing one raises a prompt asking you to confirm you trust it; if the picker is empty, that prompt is usually still waiting.
MCP Cash Book in detail · The same server in Cursor and Windsurf · Every server in VS Code · Guides · VS Code docs · Buy Pro