A comparison of agent-tool governance under the Model Context Protocol, comparing four approaches: hyperscaler agent platforms, AI-control-plane SaaS, DIY MCP servers and sovereign workspace-integrated MCP gateways.
Data plane jurisdiction, auth model, audit log unification, multi-vendor agent compatibility, deployment model. Each family hits a different point in the sovereignty/convenience trade-off.
Sovereign, on customer infrastructure, multi-agent compatible (Claude, ChatGPT, Copilot, Cursor) under one platform identity and one audit log.