Skip to content

Connecting Integrations

Connect the tools and data an agent is allowed to reach, over MCP.

How MCP connectors work

MCP is the Model Context Protocol. A connector exposes a provider's tools to an agent, and AgentZ governs every call through one policy edge rather than per agent.

Adding a connector

Pick a provider, authorize once, and wait for it to go Ready.

The OAuth flow

Coming soon

This section will cover the click path, end to end.

Callback URL

Coming soon

This section will cover what you register with the provider.

Scopes

Coming soon

This section will cover how you choose the narrowest set that works.

Tool permissions

Permission is set for each individual tool call, not for the connector as a whole. Read and scan can pass while mutate, push and delete stay denied, so a prompt injection cannot turn a lookup into a teardown.

An agent wants to call a tool, and permission is checked for this exact call. Always Allow runs unattended. Needs Approval waits for a human to confirm, then runs if approved and blocks if denied. Blocked never leaves the sandbox. Both the run and the block land in the trace.

Every branch ends in the trace, including the blocked one. A denied call is evidence, not a silent no-op.

Blocked

The call never leaves the sandbox.

Needs Approval

A human confirms before the call runs. Most teams want this for anything that writes.

Always Allow

The call runs unattended, and lands in the trace.

Credential matching and host-based auto-detection

Coming soon

This section will cover how AgentZ picks the right credential for an outbound host.

Troubleshooting common connector errors

Coming soon

This section will cover the errors teams actually hit, and the fix for each one.