Discover remote MCP servers and observed tool usage with Global Secure Access
Implementation Effort: Medium - Requires supported traffic acquisition, inspection where applicable, and access to the relevant logs.
User Impact: Low - Discovery observes traffic without adding a blocking policy; new forwarding or inspection deployments are separate changes.
Overview
Global Secure Access (GSA) Model Context Protocol (MCP) logging identifies remote MCP traffic, including previously unknown servers. It distinguishes capabilities advertised by a server from tools an agent actually invokes.
This supports Verify explicitly through observed server and tool activity and Assume breach by exposing unreviewed connections. The findings inform approval and least-privilege decisions; logging itself does not enforce those decisions.
Prerequisites
- The applicable Internet Access licensing and a supported acquisition path.
- For end-user devices: the GSA client, Internet Access forwarding, and Transport Layer Security (TLS) inspection.
- For Copilot Studio: the dedicated GSA integration, which logs this traffic without separately requiring TLS inspection.
- Global Secure Access Administrator access for configuration and the documented Global Secure Access Log Reader permissions for log access.
- Privacy and evidence-handling arrangements for potentially sensitive tool payloads.
Limitations
- Only remote MCP traffic captured through a supported route is visible. Local stdio servers need complementary Defender local-agent discovery for their configuration view.
initializeandtools/listevents describe advertised capabilities;tools/callevents describe invocations.- Destination URL is the reliable server identifier. A logged user or display name alone does not establish a unique agent identity.
- Missing events do not prove that no activity occurred. Dashboard availability does not prove collection coverage.
- Discovery does not block servers or tools; enforcement requires separate policies and an appropriate supported path.