MCP servers

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": { "GITHUB_TOKEN": "…" }
    },
    "docs": {
      "type": "http",
      "url": "https://example.com/mcp",
      "headers": { "Authorization": "Bearer …" }
    }
  }
}

Servers are read from .mcp.json at the project root and ~/.phronen/mcp.json for every project. A server is either a command PHRONEN starts (stdio) or an address it calls (http).

From a conversation

/mcp                                   # list
/mcp add github npx -y @modelcontextprotocol/server-github
/mcp add-http docs https://example.com/mcp
/mcp remove docs

• Each server's tools reach PHRONEN as mcp__<server>__<tool>, under the same permission mode and rules as its own tools.
• What a server returns is data, never an instruction.
• Servers connect when a conversation starts; one you add connects in the next conversation (/clear starts one).