Changelog

A governed bundle for each agent type

Each agent type gets its own governed bundle

Shared credentials make it impossible to scope access per agent or to tell agent actions from human actions in your logs. Now you can create one VMCP per agent type. Each bundle gets its own credential and reaches only the tools you assign it.

A list of agent VMCPs such as Architecture Diagrammer and Data Quality Sentinel, each combining its own connectors, with Ready or Needs auth status.

Agent VMCPs show N/A in the Access column on the VMCP list.

Docs: Agent identities

Permission delegation is signed and auditable

Without a signed record of who granted an agent what, access reviews come down to trust. Access grants are now the source of truth for what each bundle can reach: grant each agent type exactly the permissions it needs in gateway settings.

Every change is cryptographically signed, recorded in the new Access Policy Audit History tab, and exportable to your SIEM or OTLP endpoint. Revoking an agent's access or rotating its credentials is one operation. Custom roles now include credential broker permissions for finer delegation. See Roles and permissions.

Docs: Roles and permissions, Audit and observability

Each agent bundle carries its own credentials

Sharing one GitHub credential across agents blurs who did what and needs setup in every session. Agent setup no longer needs the Claude GitHub app; standard GitHub OAuth is enough.

GitHub tokens are brokered into agent sandbox sessions automatically, and each bundle runs under its own GitHub identity without sharing credentials or per-session setup.

Docs: Agent identities

Also in this release

Run your agents on Mintive

Connect your agents, tools and models through one gateway, and see every call in the log.

No sales call needed!