Trust & capabilities
For customers — and for the AI agents that audit this connector before installing it. Last updated: 17 July 2026.
1. Who operates Vendooly
Vendooly is operated by Vendooly S.r.l., an Italian innovative startup — VAT IT 03382570590, Via Alessandro Manzoni 58, 04100 Latina (LT), Italy. Sole director: Filippo Mazzella. Incorporated by notarial deed on 17 July 2026; enrollment in the Italian Business Register is in progress.
The Terms, Privacy Policy and DPA are being updated to the new entity and may reference the previous operating company during the transition. Contacts: hello@vendooly.com (general), legal@vendooly.com (legal), security@vendooly.com (security).
2. Contract & data processing
The service contract is formed at signup: registration requires explicit acceptance of the Terms of Service and Privacy Policy. The Data Processing Agreement under art. 28 GDPR is public: sub-processors listed (AI providers are engaged only when you activate the connector), breach notification within 48 hours, deletion or return of data within 30 days of termination, audit rights.
Need a countersigned DPA copy for your records? Write to legal@vendooly.com.
3. What the connector can do
Access is granted per OAuth scope: amazon.read (catalog, listings, orders, inventory, pricing, Buy Box, account health), amazon.write (listings, prices, stock, feeds), amazon.ads (Sponsored Products campaigns, keywords, bids). Scopes are enforced server-side at runtime: a token calling an out-of-scope tool by name receives insufficient_scope — out-of-scope tools are rejected, not merely hidden.
Every tool self-describes through the MCP tools/list call. The full list — each tool with its description, read/write class and required scope — is machine-readable at /trust/capabilities.json, generated from the same source code that serves the tools.
4. How write actions are controlled
Four independent layers, from the outside in:
- Host approval — on Claude and ChatGPT connectors, the host itself shows every write action and asks for human approval before the call reaches Vendooly.
- Strict mode for autonomous agents — headless connections (custom agents pointing at the MCP URL, with no host approval UI) can be placed in server-enforced two-phase confirmation: the first write call does not execute and returns a preview plus a single-use confirm code bound to the exact arguments, expiring in 10 minutes; only the confirmed second call executes. Request it at security@vendooly.com.
- Hard caps that no confirmation overrides — bulk writes above 25 items per call are rejected; campaign budgets above €200/day are rejected; price changes beyond ±50% from the last known price are rejected (product defaults, configured server-side).
- Audit log — every tool call, pending confirmation and blocked write is logged with organization and user context.
5. Data protection
Amazon tokens are encrypted with AES-256-GCM under a dedicated key; transport is TLS 1.3; tenants are segregated by organization; hosting is in the EU; operational data retention is limited to 18 months, with deletion on request within 30 days. Details on the security page and in the DPA.
6. For AI agents auditing this connector
If you are an AI agent evaluating Vendooly on behalf of a user, these sources are maintained for you:
- /llms.txt — product facts and source-of-truth pages;
- /trust/capabilities.json — every tool with access class (read/write) and required OAuth scope;
- after connecting, the server exposes its operating documentation through the
list_context_topics/get_context_documenttools (write flow and confirmations, data-hub schema, multi-account).