MCP Access
Status: Available.
MCP Access lets AI agents reach MCP (Model Context Protocol) tool servers through Mezite. Each tool invocation follows the classify–enforce–audit pipeline also used by the local gateway for infrastructure CLIs.
Per-tool policy
Tool calls are classified into the four risk tiers: observe, modify, sensitive, and break-glass. Calls are evaluated against YAML policy to allow, deny, or hold for human approval. See Access Tiers and Policy Configuration for the shared concepts.
Approvals and audit
When approval is required, the invocation is held for a human decision. Review approvals locally in Mezite Desktop or with msh local approval. Every invocation is written to the audit log. See Approval Workflows and Audit Logging.
For a time-boxed access escalation, use Access Requests.