AI & MCP
AI agents are privileged users. Treat them that way.
12Port’s MCP server fronts every privileged action. AI agents request access through the same broker as humans, with the same identity, the same audit trail, and the same controls. Plus six AI-native capabilities for analysts, approvers, and incident response.
Secure every AI credential with policy-driven access and full audit visibility.
How it works
An MCP server for every privileged action.
12Port speaks Model Context Protocol natively. AI agents authenticate to the MCP server with their own identity, request specific privileged actions, get vault credentials brokered through the same access broker humans use, and produce the same recorded session. The agent never holds the credential. The agent cannot bypass the policy.
Six AI capabilities also run on top of the session corpus, pulling signal forward for analysts and approvers without bolting on a separate analytics stack.
Six AI capabilities, one platform
Built for both AI agents and the humans who supervise them.
MCP Server
Native Model Context Protocol endpoint. AI agents authenticate, request privileged actions, and run them through the broker. Same identity, same audit, same policy as humans.
Ask AI
Plain-language search across every privileged session. “Show me sessions where someone disabled audit logs.” Returns ranked sessions with synchronized video and transcript.
Session intelligence
Per-session risk scoring, anomaly detection, and behavioral baselines. Surfaces high-risk sessions automatically; feeds incident response with full evidence.
Classification intelligence
Auto-tags sessions, targets, and commands by sensitivity. Compliance reports build themselves; high-risk activity surfaces without hand-curating evidence.
Intelligent MFA
Adaptive MFA enforcement based on behavior, target sensitivity, and session signals. Sessions pause for re-verification mid-flight when policy or analytics demand it.
Network intelligence
Maps lateral movement across privileged sessions. Spots when a single operator pivots through three jump hosts; visualizes blast radius before it becomes incident.
Non-human identities
How 12Port treats AI agents and automation.
First-class identity
Each AI agent gets its own identity, scopes, and audit trail. No shared service accounts. No agent piggybacking on a human’s privileged session.
Same broker as humans
No parallel access path. Agents go through the same broker, the same vault, the same approval workflows, the same recording. One control plane covers both.
Per-action approval
Configure which actions an agent can perform autonomously, which require human approval mid-loop, and which are blocked entirely. Per-target, per-role, per-scope.
Credential never disclosed
The agent gets the result of the privileged action, not the credential. Vault contents stay in the vault, even when an agent is operating against a target.
Full session evidence
Every agent session produces the same recording artifact as a human session: command transcript, file transfers, target events. Agent behavior is auditable like everything else.
