Security at Voozeo MCP
Updated October 30, 2026
Voozeo MCP is built so that your API secrets stay encrypted, your workspace stays isolated from others and your content is used only to run your servers. This page states what the service does today, without promises beyond that.
Hosting in the European Union
The service runs on Amazon Web Services in the Ireland region (eu-west-1). AWS is the only processor of the hosting. All traffic to the website, the application, the API and the MCP servers goes over HTTPS.
Secrets are encrypted
The secret of your target API (key, token or password) is written once and encrypted with a dedicated key from the AWS Key Management Service. It is never sent back by the API and never displayed again; the interface only shows that it is configured. The AI never sees it.
Your MCP keys and sign-in tokens are stored as hashes, not in clear. Passwords are hashed with Argon2id. Application secrets live in AWS secret stores, never in the code.
No AI model reads your content
Voozeo does not send your documents, tool arguments or results to any language model. The service retrieves passages and calls your API; the AI you connect is the only model involved, and it is yours. Your content is not used to train a model.
Protection against server-side request forgery
Every address Voozeo calls for you, whether to import an OpenAPI file or to run a tool, must use HTTPS and is checked after DNS resolution. Private, loopback, link-local and other reserved addresses are refused, redirections are re-checked at each step, and each request has a time limit and a size limit. The consequence is that your API must be reachable over the public internet.
Isolated workspaces
Each account has its own workspace. A server, a source or a log line of another workspace cannot be read or changed from yours: such a request is answered as if the resource did not exist. Administration actions are limited to Voozeo's super-administrators, who must sign in with a second factor.
What is logged, and what is not
The call log records which tool was called, the outcome, the HTTP status of your API and the duration. By default it never contains the arguments or the results of a call, your MCP key or your API secrets. The diagnostic mode is off by default and is enabled server by server: it keeps a masked copy of the content of the calls for a limited time, and headers and secrets are never kept. See the call log.
The call log is kept for a duration that depends on your plan, then deleted automatically. The audit trail of your workspace is kept for a limited time, then expires automatically. The cloud logs of the service never contain keys, secrets, arguments or results.
Your data, your control
From the account screen you can export your data as a JSON file and delete your account. Deletion removes your user, your workspace, your servers, their documents, your tokens and your logs. The audit entries are anonymized and then expire. See account and data.
The service does not use advertising or third-party analytics, and sets no tracking cookie. Audience measurement is aggregated and stores no address.
Report a vulnerability
If you think you found a security problem, tell us before making it public. The security contact and its validity date are published in the security.txt file, and you can also write to us from the contact page. We read every report.