AnyTool

v1.0.0Coding Agentsstable

'AnyTool: Universal Tool-Use Layer for AI Agents'

ai-agentscomputer-use-agentsmcptool-orchestrationtool-use
Share:
652
Stars
0
Downloads
0
Weekly
0/5

What is AnyTool?

AnyTool is a Model Context Protocol (MCP) server that allows AI assistants like Claude, Cursor, and VS Code to 'anytool: universal tool-use layer for ai agents'

'AnyTool: Universal Tool-Use Layer for AI Agents'

This server falls under the Coding Agents category on MCPgee, the world's largest MCP server directory with 33,000+ servers.

Features

  • 'AnyTool: Universal Tool-Use Layer for AI Agents'

Use Cases

Provide a universal tool-use layer for AI agents to orchestrate multiple tools. Enable computer-use agents to interact with any application. Build flexible, composable tool execution pipelines.
HKUDS

Maintainer

LicenseMIT
Languagepython
Versionv1.0.0
UpdatedMay 20, 2026
Statushealthy
Maintenanceactive

Works with

ClaudeOpenAIwindowsmacoslinux

Installation

Manual Installation

npx anytool

Configuration

Configuration Details

Config File

claude_desktop_config.json

Performance

Response Metrics

Response Time< 200ms
ThroughputMedium

Resource Usage

Memory UsageLow
CPU UsageLow

How to Set Up and Use AnyTool

AnyTool is a Python framework that acts as a universal tool-use layer for AI agents, solving three key challenges: overwhelming tool contexts, unreliable tools, and limited capability coverage. It provides lightning-fast tool retrieval through multi-stage progressive filtering, self-evolving orchestration that learns from past executions, and a multi-backend architecture that covers web APIs, shell commands, GUI automation, and MCP servers through a single unified interface. One async context manager call is all that is needed to give any AI agent access to a large, well-managed tool ecosystem.

Prerequisites

  • Python 3.12 or higher
  • Conda (recommended for environment management, as it includes ffmpeg for GUI recording)
  • An OpenAI-compatible LLM API key (or a locally running model) for the agent intelligence layer
  • Git for cloning the repository
  • Optional: Platform-specific GUI dependencies (pyobjc on macOS, python-xlib on Linux, pywinauto on Windows) if using the GUI or Shell backends
1

Clone the repository and create the conda environment

Clone AnyTool from GitHub and create the recommended conda environment. The conda environment includes ffmpeg which is needed for the optional video recording feature.

git clone https://github.com/HKUDS/AnyTool.git
cd AnyTool
conda create -n anytool python=3.12 ffmpeg -c conda-forge -y
conda activate anytool
pip install -r requirements.txt
2

Create a .env file with your API keys

Copy the example environment file and fill in your API keys. AnyTool supports OpenAI-compatible endpoints, so you can use OpenAI, a local Ollama instance, or any compatible provider.

cp anytool/.env.example anytool/.env
# Edit anytool/.env and add your keys
3

Choose local or server execution mode

AnyTool defaults to local mode where Shell and GUI operations run in-process — no server is needed. For remote VM control or process isolation, set mode to 'server' in anytool/config/config_grounding.json and start the local server separately.

# Local mode (default) - no additional setup needed
# Server mode - start the local server:
python -m anytool.local_server.main
4

Use AnyTool in your Python agent code

Import AnyTool and use it as an async context manager. Pass a natural language task description to execute() and the framework handles tool selection, execution, and result aggregation.

import asyncio
from anytool import AnyTool

async def main():
    async with AnyTool() as tool_layer:
        result = await tool_layer.execute(
            "Research trending AI tools on GitHub and create a comparison report"
        )
    print(result)

asyncio.run(main())
5

Install platform-specific dependencies for GUI or Shell backends (optional)

If you need GUI automation or advanced shell execution in server mode, install the platform-specific packages for your operating system.

# macOS
pip install pyobjc-core pyobjc-framework-cocoa pyobjc-framework-quartz atomacos

# Linux
pip install python-xlib pyatspi numpy
sudo apt install at-spi2-core python3-tk scrot

# Windows
pip install pywinauto pywin32 PyGetWindow

AnyTool Examples

Client configuration

AnyTool is primarily a Python SDK rather than a standalone MCP server. The following shows how to integrate it as an MCP server in Claude Desktop by running it as a Python module.

{
  "mcpServers": {
    "anytool": {
      "command": "python",
      "args": ["-m", "anytool.mcp_server"],
      "cwd": "/absolute/path/to/AnyTool"
    }
  }
}

Prompts to try

Example tasks you can pass to AnyTool's execute() method or to an MCP client when the server is running.

- "Search GitHub for the top Python MCP server libraries and compare their features"
- "Open a browser, navigate to news.ycombinator.com, and summarize the top 5 stories"
- "Run the test suite in /path/to/project and report any failures"
- "Research trending AI coding tools from GitHub and tech news, collect features, and create a comparison report"
- "Monitor the CPU usage of process X every 30 seconds and alert if it exceeds 80%"

Troubleshooting AnyTool

ImportError or missing module errors when running AnyTool

Make sure you are running inside the conda environment: 'conda activate anytool'. Then verify all dependencies installed correctly by running 'pip install -r requirements.txt' again. Some optional backend packages require platform-specific extras.

GUI automation fails with permission errors on macOS

macOS requires Accessibility and Screen Recording permissions for GUI automation. Run the local server once and grant the permissions when prompted, or manually grant them in System Settings > Privacy & Security > Accessibility and Screen Recording.

Tool execution is slow on first run

AnyTool pre-computes embeddings for all available tools on first initialization. This one-time indexing step can take a minute or two depending on the number of tools. Subsequent runs use cached embeddings and are significantly faster.

Frequently Asked Questions about AnyTool

What is AnyTool?

AnyTool is a Model Context Protocol (MCP) server that 'anytool: universal tool-use layer for ai agents' It connects AI assistants to external tools and data sources through a standardized interface.

How do I install AnyTool?

Follow the installation instructions on the AnyTool GitHub repository. Clone the repo, install dependencies, and add the server config to your AI client.

Which AI clients work with AnyTool?

AnyTool works with all major MCP-compatible AI clients including Claude Desktop, Claude Code, Cursor, VS Code (GitHub Copilot), Windsurf, and Cline.

Is AnyTool free to use?

Yes, AnyTool is open source and available under the MIT license. You can use it freely in both personal and commercial projects.

Browse More Coding Agents MCP Servers

Explore all coding agents servers available in the MCPgee directory. Each server includes setup guides for Claude, Cursor, and VS Code.

Quick Config Preview

{ "mcpServers": { "anytool": { "command": "npx", "args": ["-y", "anytool"] } } }

Add this to your claude_desktop_config.json or .cursor/mcp.json

Read the full setup guide →

Ready to use AnyTool?

Browse our complete directory of 33,000+ MCP servers, read setup guides for your editor, and start building with the Model Context Protocol.

33,000+ ServersFree & Open SourceStep-by-Step Guides