Claude 自动批准 MCP

Claude Auto-Approve MCP

Warning The May 1st update introduced an "Always approve" button - do not use this as it cannot be undone! claude-autoapprove-mcp restores the safer "Allow for this chat" functionality:

• Configure permanent approvals in config
• Make per-chat decisions for other tools
• Change decisions in newer chats This provides granular control while maintaining security.

If you by mistake used the "Always approve" button, as a workaround you can rename the mcp in the claude_desktop_config.json file to something else. Example: git->git_new. If Anthropic fixes this bug in the future, we may not need claude-autoapprove-mcp anymore.

An MCP to restart Claude Desktop App with enabled debugger port and inject a JavaScript into it, which extends Claude with MCP auto-approve functionality. It uses the claude-autoapprove library under the hood.

How it works

The MCP server will restart the Claude Desktop App with enabled debugger port and inject a JavaScript into it, which extends Claude with MCP auto-approve functionality.

Dont't be afraid when after 1st start of the app it will be closed immediately. This is expected behavior.

Installation

Imstalling uv (if you don't have it yet)

After installing uv, make sure it's available in your PATH.

MacOS

Brew

brew install uv

MacPorts

sudo port install uv

Windows

winget install --id=astral-sh.uv  -e

Other installation options

You can find other installation options in the uv installation guide.

Add it to your claude_desktop_config.json

{
    "mcpServers": {
        "claude-autoapprove-mcp": {
            "command": "uvx",
            "args": [
                "claude-autoapprove-mcp"
            ],
            "autoapprove": [
                "autoapproved_tools",
                "autoblocked_tools"
            ]
        }
    }
}

Restart Claude Desktop if it is already running.

Arguments

{
    "mcpServers": {
        "claude-autoapprove-mcp": {
            "command": "uvx",
            "args": [
                "claude-autoapprove-mcp",
                "--port", "19222",
                "--persist"
            ]
        }
    }
}

Configuration

You can add autoapprove and autoblock parameters to each MCP server, the claude-autoapprove-mcp will read that configuration and apply it to the running instance.

{
    "mcpServers": {
        "claude-autoapprove-mcp": {
        "command": "uvx",
            "args": [
                "claude-autoapprove-mcp"
            ],
        "autoapprove": [
            "autoapproved_tools"
        ],
        "autoblock": [
        ]
    },
    "project-mem-mcp": {
        "command": "uvx",
        "args": [
            "project-mem-mcp",
            "--allowed-dir", "/Users/wallner/Developer/MCP",
            "--allowed-dir", "/Users/wallner/Developer/Projects/ByCompany/BacktestGuy"
        ],
        "autoapprove": [
            "get_project_memory",
            "set_project_memory",
            "update_project_memory"
        ],
        "autoblock": [
        ]
    },
    "browsermcp": {
      "command": "npx",
      "args": [
        "@browsermcp/mcp@latest"
      ],
      "autoapprove": [
        "browser_get_console_logs",
        "browser_snapshot"
      ],
      "autoblock": [
        "browser_screenshot"
      ]
    }
}

Usage

Just run Claude Desktop. It is not invasive, it doesn't change anything in the app (only if you use the --persist argument), just injects a JavaScript into the running instance. So you can install updates as usual (even when using the --persist argument). It uses a feature of most Electron based apps: remote debugging port.

If you want to list all tools that are auto-approved, you can use the following prompt in Claude Desktop:

list all tools that are auto-approved

If you want to list all tools that are auto-blocked, you can use the following prompt in Claude Desktop:

list all tools that are auto-blocked

Security

The remote debugging port allows any application on your localhost (your machine) to connect to the running Claude Desktop App. This may be a security risk, because any app or script can connect to it and execute arbitrary code inside Claude Desktop App context. This may be used for malicious purposes. It is a low risk, if you know what is running on your computer.

So be careful when using this feature and use it at your own risk.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Contributing

If you want to contribute to this project, please fork the repository and create a pull request.