AnyTool

AnyTool is a Universal Tool-Use Layer that enables AI agents to orchestrate a wide range of tools in a scalable, safe, and production-ready manner. The MCP server provided by this repository runs the AnyTool local server, which exposes an API to execute tool-using workflows remotely or locally via HTTP. You can run the server in Python mode and then have your agent communicate with it to perform shell commands, GUI actions, or other tool interactions through AnyTool's orchestration layer. This MCP server essentially enables remote or isolated execution of tool-layer tasks via a Flask-based service, making it suitable for multi-machine deployments or sandboxed environments.

To use it, start the server and point your agent to the HTTP API endpoint exposed by the local_server. The AnyTool layer supports both local in-process execution and server-based remote execution, controlled by the AnyTool configuration. When your agent calls tool_layer.execute with a descriptive prompt, AnyTool selects and orchestrates the appropriate tools, handles context management, and returns structured results. The system is designed for fast tool retrieval, scalable tool orchestration, and universal automation across backends.

Prerequisites:

• Python 3.12 or newer
• Git installed on your system
• (Optional) Conda for environment management

Steps:

1. Clone the repository:

git clone https://github.com/HKUDS/AnyTool.git
cd AnyTool

2. Create and activate a Python environment (recommended):

• Using conda:

conda create -n anytool python=3.12 ffmpeg -c conda-forge -y
conda activate anytool

• Or using venv:

python -m venv venv
source venv/bin/activate  # on macOS/Linux
venv\Scripts\activate.bat # on Windows

3. Install dependencies:

pip install -r requirements.txt

4. Start the local MCP server (AnyTool local server):

python -m anytool.local_server.main

5. Verify by sending a test request to the exposed API endpoint (the README/local_server docs provide API details).

Notes:

• If you plan to run in server mode, ensure platform-specific dependencies are installed as described in the README (macOS/Linux/Windows setup blocks).
• Create a .env file in the repository root and add your API keys as referenced in the documentation (anytool/.env.example).

Tips and notes:

• The server can run in Local or Server mode depending on your deployment needs. Local mode runs in-process, while Server mode exposes a Flask HTTP interface for remote tool execution.
• For GUI and shell automation, ensure the appropriate platform dependencies are installed (macOS/Linux/Windows sections in the README).
• If prompts for permissions appear (macOS), grant Accessibility and Screen Recording as needed for GUI tasks.
• The .env file may contain API keys and other configuration values; keep it secure and do not commit to version control.
• The MCP config shown here uses the Python module invocation to start the server; adjust env vars as needed for your environment.