CLI vs MCP
Both talk to the same plugin. Use the CLI (direct, fast, token efficient); use the MCP server only when your AI cannot run shell commands.
Both talk to the same plugin. Use the CLI (direct, fast, token efficient); use the MCP server only when your AI cannot run shell commands.
Both talk to the same plugin. The CLI is recommended: it is direct, faster, and more token efficient for AI agents. Use MCP only when your AI tool cannot run shell commands.
Your AI runs cfa commands through its shell tool. Best for Claude Code.
cfa health_checkFor Cursor, Windsurf, VS Code and other MCP clients. The same binary serves MCP over stdio.
cfa install claudecfa install <client> auto registers the MCP server. Supported clients are claude, cursor, vscode, windsurf, gemini, jetbrains, zed, amazonq, and codex (or all); add --scope project for a project-local config. To wire it by hand, add an MCP server with command cfa and args ["mcp"].