Skip to main content

Access Control

Restrict which AI agents can use each MCP server.

Overview​

Access Control lets you restrict each MCP to specific AI agents. When enabled, only the selected agents can invoke tools from that MCP. Agents are identified by their User-Agent header keyword.

How It Works​

  • Per-MCP agent-level access control - Configure allowed agents for each MCP independently
  • Toggle access for predefined agents - OpenAI, Claude, Cursor, Gemini
  • Support for custom agents - User-Agent keyword matching for any AI client
  • Changes take effect immediately - No restart needed

Predefined Agents​

Pre-defined agents (OpenAI, Claude, Cursor, Gemini) are available by default. Each is identified by its User-Agent header keyword.

Custom Agents​

Create custom agents with a keyword and display name to support any AI client. The gateway matches the keyword against the User-Agent header of incoming requests.

Configuration​

Access control is configured per-MCP in the Settings → General tab. Toggle individual agents on or off for each MCP server.

tip

For a global view of agent permissions across all MCPs, use Agents Configuration instead. Both views control the same underlying permissions.