AI agents and MCP

Your teams use AI agents, such as Claude Code, Cursor or an agent you built, to call tools like Google Drive, Jira or your own APIs. PolicyArc gives each agent its own identity and puts the managed gateway between the agents and those tools. Policy decides each tool call, and each decision is recorded. The gateway allows or refuses a tool call. It never holds one for a person. Approval happens earlier, when an agent asks /token for a token that a policy rule holds, and a person then approves it through CIBA.

The agent access problem tells the whole story, with what PolicyArc does about each gap.

Try it

• First environment sets up an environment ready for a first governed call.
• Connect clients connects an MCP client such as Claude Code.
• Hold an agent's write for a person's approval runs an agent, a held write and an approval on your environment.
• Connectors lists the tools the gateway can front.