AI Access & MCP

AI Access separates authentication from authorization.

supported OAuth / credential
→ verified AI Identity
→ AI Access policy
→ Permission Object / Group
→ approved target

A display name alone is never authentication.

AI Identity

set ai-identity <NAME>
show ai-identities
show ai-identity <NAME>

Interactive AI uses OAuth Authorization Code verification where supported. Automation/custom AI can use OAuth Client Credentials where supported.

Permission Objects

Examples include host-info, process-read, file-read, command-exec, file-write, file-upload, and file-download. A true read-only permission set does not include command execution.

AI Access Rule

set permission-object read-only permissions host-info,process-read,file-read
set ai-access claude-prod mode whitelist source claude destination production-servers permission read-only enabled
test ai-access source claude destination production-servers permission read-only

AI Access Rules are not ordered and have no per-rule action. AI authentication remains mandatory even when policy enforcement is disabled. The v2.4 target uses a server-side MCP Bridge over authenticated HTTPS. MCP does not bypass Agent identity, AI Access policy, or target OS permissions. Remote Services and SSH are not prerequisites for approved AI operations.

Last modified on September 22, 2026

Getting Started

Managed Hosts & Objects

Remote Services & Access

Internet Access

AI Access

CLI & Configuration

Deployment

Operations

Security & Reference

AI Access & MCP

AI Access & MCP

AI Identity

Permission Objects

AI Access Rule

​AI Access & MCP

​AI Identity

​Permission Objects

​AI Access Rule

AI Access & MCP

AI Identity

Permission Objects

AI Access Rule