MCP servers

MCP (Model Context Protocol) servers give Xeneric's AI new tools: GitHub, a database, your files, anything with an MCP server. Manage them in Dashboard → MCP.

Add a server#

1. Click Add server

   Choose a program on this computer, or a server at a URL.

2. Fill it in

   For a program: a name, the command (for example npx -y @modelcontextprotocol/server-filesystem ~/Documents), optional environment variables (GITHUB_TOKEN=…) and folder. For a URL: the address (for example https://mcp.example.com/mcp) and optional headers (Authorization: Bearer YOUR_TOKEN).

3. Turn on “Let Xeneric use these tools”

   Until then, the AI doesn't see the server.

Programs talk over stdin/stdout; servers at a URL use Streamable HTTP. A server with the same name, command or URL as one you've added is refused.

Paste MCP JSON#

Paste JSON takes the mcpServers config from Claude Desktop, Cursor or a server's README:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/you/Documents"]
    }
  }
}

Import from Claude Code#

Xeneric finds the servers you set up in Claude Code (in ~/.claude.json and projects' .mcp.json files) and offers to import them. Imported servers stay off until you turn them on.

What the AI may use#

Let Xeneric use these tools

Turns a server on for the AI. Tools that only read are offered.

Allow actions

Also offers tools that create, send, change or delete. Off by default.

There's no per-call confirmation in the chat yet, so only allow actions for servers you trust. Offline mode stops servers at a URL; programs on your Mac keep working.

Where secrets go#

Environment values and header values never go in settings.json. Xeneric keeps only their names there; the values live in the macOS Keychain.