VibeCFO vs a DIY build: Claude, an MCP and a weekend

What does a DIY build actually do well?

One org, one simple question, fast, cheap, no procurement. A Claude subscription and a ledger MCP will answer a straightforward question about a single organisation in seconds, and for that use case it genuinely works. If that is the scope of the problem, use it.

Where does it stop on depth?

A raw MCP returns what the API hands back easily - trial balance, P&L, aged receivables. Summaries. That answers the first question: what was revenue last month. It cannot answer why, because the transactions underneath that summary were never pulled. Invoice lines, journal entries, timesheet rows, years of history - that is the layer where the second question lives, and it is not there.

Where does it stop on breadth?

One connector sees one system. WIP sits in simPRO. Revenue sits in Xero. The client record sits in XPM. Same customer, three different names for them. A blended answer needs the same customer resolved across all of them, and a single connector cannot do that because it can only ever see itself.

What happens at the second question?

The DIY build handles turn one. Turn three needs data that was never pulled - the invoice lines underneath the summary. Turn four needs a system that was never connected - the job management platform where the WIP actually sits. Each follow-up is a new prompt and often a new script, not a continuation of a conversation against joined data.

What happens across every client?

A builder loops until one answer looks right on one org. That is one answer, checked once, by someone who knew what they were looking for. Correct on the org you hand-checked is not correct across the two hundred you did not. Each one has a different chart of accounts, a different stack, its own consolidations and intercompany eliminations. That gap is where every weekend build quietly dies.

Is hand-checking the same as testing?

No. A builder checks the output looks plausible, once, by eye. Our standard is right the first time, on questions nobody has asked yet, with nobody squinting at the output. That is the difference between a checked answer and a tested system.

What would it take to close the gap?

Rebuild every integration at transaction depth. Entity resolution across them. Multi-tenancy. Reconciliation on every refresh. Compliance. Support. Then maintain all of it while the APIs change underneath you. That is not a weekend project. It is years of engineering and it is what VibeCFO is.

The second question, shown

1

“How did we go last month?”

Xero

Revenue $312K, up 8% on prior month. Margin dropped from 34% to 29%.

2

“Why is margin down?”

Xero

Three job types carrying the drop. Fixed-price installs down 11 points.

3

“Which jobs specifically?”

XerosimPRO

Four jobs running over quote. Two billed under actual hours by 40%+.

4

“Is any of that still sitting in WIP?”

simPROXPM

$47K WIP unbilled across those clients. $18K older than 60 days.

5

“Now do that for every client on the books.”

All connected

Same analysis, every entity. 14 clients with WIP older than 30 days.