MCP servers come in two shapes: local servers launched as subprocesses from a config file, and remote servers reached over HTTP, often added as connectors in chat products. Both widen what an agent can read and do. Each needs an inventory entry with its transport, command or URL, auth method and owner.
Read in this order
- OfficialMCP: Connect to local MCP serversStart here: how a local server is launched and what config it needs.
- OfficialMCP: Connect to remote MCP serversThe remote case: HTTP transport, URLs and the auth that comes with them.
- VideoShadow MCP Servers: The AI Security Risk Hiding on Your Developers' Laptops (The Application Security Podcast)Explains the inventory gaps left by developer-installed MCP servers. 47 min.