Clawvisor

Clawvisor is an open-core security gateway and control plane for AI agents. It sits between your agents and the tools they interact with—Claude Code, MCP servers, Gmail, GitHub, Slack, databases, APIs—and replaces the bad tradeoff between speed and safety. Instead of giving agents standing credentials or requiring per-call approval, Clawvisor lets you approve a task once. Every subsequent tool call is verified against the task's scope, scoped credentials are swapped in at call time, and a full audit trail is recorded. Built for engineering and security teams running agents in production, Clawvisor offers fail-closed containment (agents start with zero access), a credential vault where real secrets never reach the agent, policy in plain English (RBAC/ABAC rules), per-task cost attribution, and risk scoring on every task. The open-source core can be self-hosted with no seat caps; the managed cloud version adds team governance, extended retention, and priority support. Key features include task-level authorization—the core innovation—where access binds to the declared purpose of a task. Combined with risk scoring, credential isolation, and comprehensive logging, it provides defense in depth without slowing agents down. The code is open and auditable, emphasizing transparency over trust. Integrations span Gmail, Google Calendar, GitHub, Slack, Notion, Linear, Postgres, Stripe, Dropbox, Filesystem, Shell, and MCP servers. Compared to alternatives like HumanLayer (per-call approval), Arcade (standing permissions), or Portkey (LLM gateway), Clawvisor's task-scoped model uniquely balances speed and safety—approve once, let the agent run, and stay in control.