airlock

Airlock is a governance tool for MCP, Skills, and Agents. It is built to help teams apply rules across AI tools and keep audit records as they move between clients, vendors, and workflows. The site describes it as a single connector for Claude, ChatGPT, Cursor, Copilot, Gemini, and other clients.

Its main functions include setting policy server-side once and applying it across tools and teammates. The listed controls include approvals, an audit log, a kill switch, and a credential vault. Airlock also mentions per-tool policy tags for read-only or destructive actions, HITL approvals through Slack, Teams, or a dashboard, rate limits and quotas per tool, team, or key, and pipeline checks for prompt-injection scanning plus secret and PII redaction. Calls are logged and can be exported to SIEM systems.

The product also treats prompts or workflows as versioned, verified Agents or Skills. It says these objects can be built once and auto-injected into Claude, ChatGPT, Cursor, Copilot, or other tools, with diff, rollback, and attestation support. Another stated feature is organization memory, where entities, decisions, and context accumulate in airlock rather than staying inside a single vendor. The page also shows an MCP catalog, pending approvals, and an agent and skill library in its Control Room.

Airlock is described as a cross-tool proxy, with one URL fronting any MCP-compatible client, and it says switching or adding a tool does not require re-onboarding. The page names client-facing teams, agencies, teams working with clients, and teams running AI trainings as intended users, and it says separate tenants can be set up per engagement. Pricing includes a free start option and team pricing. The page also says the product is in a limited beta and is backed by Airlock BV.