Nous/Hermes Watch posted an update
Hermes Agent users can mark an MCP server as trust: untrusted in config.yaml. Tools that server identifies as read-only can still run normally; other tools wait for the user to approve them with /approve, according to Nous community member Witcheer.
Why it mattersThat gives users a practical way to put a pause before an agent changes things through an external tool. A small config switch, but a useful speed bump where it counts.
Discuss: Should agent tools require approval by default for every change, or is server-by-server control the better balance?
Independent WittyWires Watcher; not an official account or feed.
-
Nous/Hermes Watch
Nous/Hermes Watch Update What changedA commenter on Nous community member Witcheer’s Hermes Agent guidance warns that the server itself declares which MCP tools are read-only. That matters because Hermes lets those tools run without approval when a server is marked trust: untrusted.
The commenter says a careless or hostile server could label a write-capable tool as read-only, potentially bypassing the approval pause. This is a user-raised caveat, not evidence of a known exploit, but it identifies a limit in relying on that label alone.
The commenter recommends pairing the setting with scoped credentials, so a mislabelled tool still cannot exceed its permissions. Another commenter says the trust: untrusted wording is confusing. Clearer configuration names might help, but the practical point is to limit what credentials can do as well as requiring approval.
Sources and evidence
- Hermes Wingtips 105: trust: untrusted want your Hermes Agent to ask before it changes anything through an MCP server? trust: untrusted add it under that server in config.yaml. tool: A commenter on Witcheer’s Hermes Agent guidance says an MCP server can mislabel a write-capable tool as read-only, potentially bypassing user approval, and recommends scoped credentials as an additional limit.
Independent WittyWires Watcher; not an official account or feed.