MCP Servers

Model Context Protocol (MCP) is an open standard that enables AI models to interact with external tools and data sources.

aix provides first-class support for configuring MCP servers and syncing that configuration to any editor that supports MCP (Cursor, VS Code, Claude Code, Windsurf, Zed, Codex).

Configuring Servers

Define servers in the mcp object in ai.json. There are two ways to connect to an MCP server:

1. Stdio (Local Process)

Run a local command (like npx or a python script) that speaks MCP over standard input/output.

{
   "mcp": {
      "github": {
         "command": "npx",
         "args": ["-y", "@modelcontextprotocol/server-github"],
         "env": {
            "GITHUB_TOKEN": "${GITHUB_TOKEN}"
         }
      }
   }
}

2. HTTP (Streamable HTTP)

Connect to a remote MCP server over HTTP.

{
   "mcp": {
      "brave-search": {
         "url": "http://localhost:3000/mcp"
      }
   }
}

Tool & Resource Filtering

You can restrict which tools and resources an MCP server exposes to the AI.

{
   "mcp": {
      "filesystem": {
         "command": "npx",
         "args": ["-y", "@modelcontextprotocol/server-filesystem", "."],
         "tools": {
            // Only allow reading files, not writing
            "include": ["read_file", "list_directory"],
            "exclude": ["write_file", "edit_file"]
         }
      }
   }
}

Global vs. Project Config

Some editors (Windsurf, Zed, Codex) only support a global MCP configuration file, while others (Cursor, VS Code, Claude Code) support project-specific config.

aix handles this complexity for you:

• For project-specific editors, it writes to the project config.
• For global-only editors, it merges your project’s MCP config into the global config and tracks which project added which server.
• When you remove a server or delete the project, aix global cleanup keeps the global config tidy.

Finding Servers

Use aix search to browse the official MCP Registry:

aix search --type mcp

This interactive command lets you find servers (like GitHub, Slack, Postgres, Brave Search) and add them to your ai.json with one click.