Proposal: optional governed AI (MCP) access (off by default, no UI)

[sandeepshekhar26]

Hi — Books being local-first with no API is exactly why I think it'd shine with a governed AI
integration, so I wanted to float it before opening a PR.

I'd like to contribute an optional, off-by-default, backend-only way to let a user's AI assistant
operate Books safely as a governed MCP server: reads/reports flow freely, routine record-keeping is
audited, and creating a financial document or submitting one (which posts the double-entry ledger) —
plus anything destructive — pauses for one-click human approval, all signed into an audit log.

It adds no dependency, no UI, and no change to the app: the handler reuses the same headless fyo
layer your test harness uses (new Fyo({ DatabaseDemux: DatabaseManager, isElectron: false })), so an
agent call runs the identical Doc.sync()/.submit() a human action does. It's a self-contained
kriya-mcp/ folder (+ a one-line .eslintrc.js ignore), inert unless explicitly enabled.

I've built and verified it end-to-end (create party → item → invoice → submit posts a balanced
ledger). Would a PR be welcome? Glad to scope it however you prefer.

[sandeepshekhar26]

#1544 (comment) changes are ready here. Let me know if you need a demo as well.