Skip to main content
Xerg graphs MCP configuration alongside your apps and services: Configured agent → MCP server → recognized service For example, a repository’s Cursor configuration can point to a GitHub MCP server. The graph keeps the configuration path and the hostname, known package or environment variable name that identifies GitHub. The server still says Access not verified, even if someone has connected a GitHub account to Xerg.

What is detected

Xerg reads at most 100 server entries per configuration file, within the normal 400 KB configuration-file limit. Only configuration inside the scanned repository is read. Home-directory settings and live agent sessions are not inspected. A server with an unrecognized endpoint or package still appears; Xerg only adds an underlying service when specific configuration evidence matches a service it recognizes. It does not guess from the server’s display name.

Find and inspect

  • Search by agent or server name, MCP, or the remote endpoint’s origin.
  • In Inventory, open Configured agents or MCP servers.
  • Select a server for its transport, enabled state, source configuration and links to recognized services. In the grid, selecting a node reveals its configured MCP chain.
  • In Ask Xerg or Xerg’s own MCP tools, search with kind: "agent" or kind: "mcp". get_item returns configuration evidence and access status; where_used finds variable names in MCP configs; what_changed reports changes to the saved MCP configuration; impact follows enabled configuration links and excludes disabled ones.

Detection and access are different

Access not verified means Xerg has not contacted the MCP endpoint, tested credentials, listed available tools, or checked what the agent is allowed to do. Enabled means the server is not disabled in this file. Machine-level settings, approval choices and personal overrides may still prevent its use; Xerg does not inspect them. It is not proof that the server is running. A disabled server stays on the graph as Disabled in config. Connecting the underlying vendor account lets Xerg read that account. It does not establish that a separately configured agent can reach it or use its tools. The graph can show a possible dependency; it cannot certify an agent’s effective permissions or predict whether a particular action will succeed.

What is kept

Xerg keeps configuration paths, server names, recognized package names, transport, enabled state, environment variable names, and remote endpoint origins (scheme, host and port). It discards credential values, raw command lines and arguments, HTTP headers, and URL paths, queries and fragments. Configured commands are never executed. See What Xerg reads and Security and data.