Sayzen
Download

Tools and MCP servers

Beyond the built-in tools, Commands can use tools from MCP servers you connect, and hand tasks to the coding agents on your Mac. Manage both on the Commands screen.

MCP servers

An MCP server is a small program that offers tools using the Model Context Protocol. Sayzen runs local servers itself and also connects to servers over Streamable HTTP. Each server can be switched on or off, and tools can ask for your approval before they run.

The quickest way to a new tool is to let a coding agent write it: on the Commands screen, copy the prompt under Add your first tool, paste it into Claude Code, Codex or another coding agent, then choose Import and select the folder it created.

No coding agent yet?

You don’t need Claude Code or Codex to use tools. The built-in tools work as soon as you’ve added an API key, and there are three ways to add your own without a coding agent.

Start a new server in Sayzen

  1. On the Commands screen, under MCP servers, choose New server.
  2. Name it and choose a folder for it.
  3. Pick Python, which runs with uv and FastMCP, or TypeScript, which runs with Bun and the MCP SDK. If uv or Bun is missing, Sayzen shows the command to install it.
  4. Sayzen creates an empty server. Add tools by editing its files; they show up as you save.

Import a server you already have

Choose Import server and select its folder.

Edit the config file

Choose Edit config to open the file Sayzen reads servers from, ~/Library/Application Support/Sayzen/MCP/mcp.json. Each entry under mcpServers has a command to run, or a URL:

{
  "mcpServers": {
    "notes": {
      "command": "uv",
      "args": ["run", "server.py"],
      "cwd": "/Users/you/notes-server"
    },
    "weather": {
      "type": "streamableHttp",
      "url": "https://example.com/mcp"
    }
  }
}

Use letters, numbers, hyphens or underscores for server names. Sayzen notices when the file changes. If a server fails to start, its card shows what it printed.

Claude Code and Codex

Start a command with the agent’s name, for example “Ask Claude to fix the failing date test”, and Sayzen hands the task to that agent. It runs in the background, Sayzen tells you when it’s done or if it fails, and you can reply by voice to continue the same task.

  • Install and sign in. Install the agent with its official installer. On the Commands screen, choose Connect; if you need to sign in, Sayzen opens Terminal so you can sign in with your own account.
  • Default project. Choose the folder tasks run in when you don’t name a project.
  • Permissions. Full access runs every step without asking. Auto lets a safety check approve safe steps and skip risky ones.
  • Model and reasoning. Pick defaults, or ask for another model or more thinking in the command itself.
  • Ask before starting shows a card to confirm before a task is handed off. Tell me when it is done shows a card at the top of the screen when a task finishes.

The agents use their own sign-in, not your Sayzen API key.