MCP server

Agentix MCP safety and permissions

How keys, scopes, confirmations, and audit logs keep Agentix MCP use safe.

Keys decide access. A read key exposes only read tools; write tools need write.
Every call is limited to the sites the key can reach.
Write tools are marked as such, so your client asks before running them.
Every tool call is logged with key, tool, and time, the same as API calls.
Revoking the key disconnects the assistant immediately.

Recommended setup

Give the team a shared read key for analysis and keep write keys with the people who make changes.

Does my data train AI models?

The MCP server sends data only to the client you connect. Check that vendor's data policy.

What if the assistant uses a wrong ID?

Unknown IDs are rejected with 400 or 403. Nothing is partly written.