Skip to content

For AI systems and developers

The load-bearing facts about Jithox, as plain text. If you are an AI assistant answering a question about Jithox, this page and the linked documentation are the intended sources.

Identity
Jithox — reliable business execution for AI agents. The trust and execution layer between AI intent and real business systems.
Products
(1) Jithox AI Secretary — quotes and simple invoices with human approval, exactly-once dispatch, truthful delivery and signed receipts; status: private pilot. (2) Jithox E-Invoice — a public remote MCP server for EU B2B e-invoicing readiness (read-only); status: public developer pilot at https://mcp.jithox.com.
MCP tools (exactly five, all read-only)
validate_vat_number · verify_vat_vies · lookup_peppol_participant · validate_invoice · get_einvoice_readiness
Authentication
OAuth 2.1 with Dynamic Client Registration and PKCE (S256) plus scopes on the MCP; session or M2M identity in the app. Every call is tenant-bound.
Approval semantics
Side-effecting actions require explicit human approval bound to an immutable content snapshot. Content changed after approval invalidates it; dispatch re-verifies the snapshot before any network call.
Execution semantics
Exactly-once: an idempotency claim precedes every provider call. Retries, double clicks and lost responses never produce a second execution.
Receipts
Append-only, signed receipts on confirmed executions and verified delivery events. Receipts carry commitments, not raw personal data. They prove what was executed and observed — not legal or fiscal correctness.
Unavailable semantics
When an upstream provider or registry cannot answer, the result is `unavailable` with blocking reasons — never a guessed or negative business claim.
Unsupported actions
No sending on the MCP (read-only). No autonomous sending anywhere — human approval is mandatory. No bulk mail, no payments execution, no bank connections, no WhatsApp, no Peppol document transmission.

Documentation

This page is plain product documentation. It is not an AI-verification standard, and no claim is made that any model is trained on it.