MCP Server Discovery - Velatir Documentation

Velatir for Desktop reads the MCP configuration of supported AI clients on each device and reports which servers are set up. You see the result in MCP servers on Insights.

Supported AI clients

AI client	Windows	macOS
Claude Desktop	✓	✓
Claude Code	✓	✓
Cursor	✓	✓
Visual Studio Code and VS Code Insiders	✓	✓
Visual Studio	✓
Windsurf	✓	✓
Zed	✓	✓
Warp	✓	✓
JetBrains Junie	✓	✓
Kiro	✓	✓
Cline	✓	✓
Roo Code	✓	✓
Continue	✓	✓
OpenCode, including OpenWork	✓	✓
OpenAI Codex	✓	✓
Gemini CLI	✓	✓
GitHub Copilot CLI	✓	✓
Amazon Q Developer	✓	✓
Goose	✓	✓

Velatir reads servers configured for the user, servers in projects the AI client already knows about, and servers your IT team manages centrally for clients that support it.

What Velatir collects

For each server, Velatir collects what identifies it:

its name and the AI client it is configured in
the package it runs from, such as an npm or PyPI package, or a container image
for a remote server, the host and path it connects to
the names of the environment variables and headers it uses

Values stay on the device: tokens, API keys, command arguments, query strings, and project folder names. If part of a remote address looks like it could hold a key, Velatir keeps only the host.

Turning it on or off

MCP server discovery is part of local AI-app detection. It runs a few times a day while that is switched on for your organisation. Switching app detection off stops new reports, and the servers from that device drop out of the list.