Feather includes a built-in Model Context Protocol (MCP) server that lets AI tools on your computer access your open documents, workspace files, and scratchpads. MCP is an open standard supported by a growing list of AI tools.

Everything runs locally on your machine. No API keys, no cloud services, no data leaves your computer.

Supported AI Tools

Any MCP-compatible client works with Feather, including:

• Claude Desktop and Claude Code
• Cursor
• VS Code Copilot (Agent mode)
• Windsurf
• JetBrains IDEs
• Zed
• Amazon Q Developer

Quick Setup

1. Open Settings > AI Integration in Feather
2. Toggle the MCP server on
3. Click Configure next to your AI tool (Claude Desktop, Cursor)
4. Restart the AI tool for the changes to take effect
5. Start a conversation in your AI tool and it will have access to Feather

For tools not listed (like Claude Code or VS Code Copilot), click Copy path under "Other AI tools" and add it to your tool's MCP configuration manually.

What AI Tools Can Do

Once connected, your AI tool can interact with Feather in these ways:

Read Your Documents

• List all open tabs with file names, paths, and word counts
• Read the full content of any open tab
• Get the heading outline of a document
• See which tab is currently active

Edit Your Documents

• Insert text at any position in a tab
• Replace a range of text in a tab
• Apply markdown formatting (bold, italic, headings, links, etc.)

Work With Your Files

• Browse workspace folders and files
• Search across all workspace files for specific text
• Open files in new tabs
• Create, rename, and delete files in workspace folders

Read Your Scratchpads

• List all scratchpads with titles and open/closed status
• Read the content of any scratchpad, including live unsaved content from open scratchpads

Manual Configuration

If you prefer to set up the connection manually, or your AI tool isn't in the auto-configure list, add Feather's MCP server to your tool's config file.

The server executable path is:

C:\Users\<your-username>\AppData\Local\Feather\feather-mcp.exe

You can copy this path from Settings > AI Integration > Other AI tools > Copy path.

Claude Desktop

Add to %APPDATA%\Claude\claude_desktop_config.json:

{
  "mcpServers": {
    "Feather": {
      "command": "C:\\Users\\<you>\\AppData\\Local\\Feather\\feather-mcp.exe"
    }
  }
}

Claude Code

Add to .mcp.json in your project root:

{
  "mcpServers": {
    "Feather": {
      "command": "C:\\Users\\<you>\\AppData\\Local\\Feather\\feather-mcp.exe"
    }
  }
}

Cursor

Add to %USERPROFILE%\.cursor\mcp.json:

{
  "mcpServers": {
    "Feather": {
      "command": "C:\\Users\\<you>\\AppData\\Local\\Feather\\feather-mcp.exe"
    }
  }
}

VS Code Copilot

Add to .vscode/mcp.json in your project. Note: MCP tools only appear when Agent mode is active.

{
  "servers": {
    "Feather": {
      "command": "C:\\Users\\<you>\\AppData\\Local\\Feather\\feather-mcp.exe"
    }
  }
}

Troubleshooting

AI tool shows "disconnected" or can't connect

• Make sure Feather is running and the MCP server is toggled on in Settings
• Check that the status indicator shows a green checkmark
• Restart the AI tool after making config changes

Tools not appearing

• VS Code Copilot: MCP tools only appear in Agent mode
• Windsurf: Has a limit of 100 tools across all MCP servers
• Make sure there are no JSON syntax errors in your config file

Server path not found

The MCP bridge executable is automatically deployed to %LOCALAPPDATA%\Feather\ when you enable the MCP server. If the file is missing, toggle the MCP server off and on again in Settings.

Privacy & Security

• Local only: All communication happens over a local pipe on your machine. Nothing is sent to the internet.
• No API keys: MCP is a local protocol. No accounts or authentication needed.
• You control access: MCP is off by default. You choose when to enable it and which AI tools to connect.
• File operations are scoped: AI tools can only access files in your workspace folders and open tabs.
• Deletions are recoverable: File deletions go to the Windows Recycle Bin.