Where this offering fits
MCP Security Governance addresses the servers and tool interfaces exposed to or consumed by AI agents. Begin with an inventory and the permissions each tool can exercise. Discovering an MCP server is not evidence that its tools are appropriate for every agent or business workflow.
Architecture and rollout decisions
Confirm current discovery, posture and testing capabilities for the proposed offer. Pilot a known server using synthetic data and bounded permissions. Review authentication, tool descriptions and the authority required for actions. Test requests outside the permitted scope and define human approval for consequential operations. Regulatory mappings can support review but do not establish compliance by themselves.
An evaluation scenario for this offering
Create a fictional MCP server in an approved test environment with one read-only tool and another consequential tool requiring explicit approval. Use synthetic records and narrow identities to verify discovery and the currently supported governance checks. Ask an agent or test client for an operation outside its permitted task and inspect both the requested action and actual authorisation outcome. Then revoke the test identity and confirm the documented effect. Retain server ownership, tool permissions, approval and logging evidence. The evaluation should distinguish knowing that a tool exists from granting authority to use it for every operation or dataset it exposes.
Evidence to request for this scope
- Identify MCP servers, owners and permitted tool actions.
- Validate access boundaries and supported security checks.
- Record approval, logging and revocation procedures.