ATaC

ATaC is a Python-based MCP tool runtime that lets agents register and reuse tools across workflows. It exposes two core capabilities: registering tools (both built-in agent tools and external MCP tools) and invoking these tools from graph code via a simple tool_call interface. The Python SDK enables graph code to call registered tools directly, enabling reusable workflow components. To use ATaC, first ensure tools are registered in AtacService, either by registering built-in tools or by exposing external MCP tools into the same tool space, after which graph nodes can call these tools by name. This setup lets you compose complex workflows by chaining tool calls inside LangGraph graphs, reusing assets and experiments across projects. For example, you can register a bash-like tool that runs shell commands and then invoke it from a LangGraph graph using get_service().tool_call("bash", {"command": "echo hello"}). The included examples show how to register a tool, how to call it from graph code, and how to use the Python SDK to trigger graph execution programmatically. The result is a modular, reusable workflow runtime that makes it easy to build, test, and reuse tool-based assets across multiple agent workflows.

Prerequisites:

• Python 3.8+ (for running the ATaC Python runtime)
• The uv tool (for installing MCP runtimes) or access to your preferred environment manager that supports uvx

Installation steps:

1. Ensure the uv tool is installed in your environment. If not, follow your platform's instructions to install uv.

2. Install the ATaC MCP server package via uvx (uses the PyPI package atac):

uvx atac

3. Verify installation by listing available MCP servers or calling a small sanity check in Python by importing ATaC and printing the registered tools (example):

uv tool list

4. (Optional) If you prefer to install and run via Python directly, you can install the package from PyPI with pip and run a minimal script to initialize a client and load a graph:

pip install atac

from atac import AtacService, set_service
service = AtacService()
set_service(service)
print('ATaC installed and service initialized')

5. To start using ATaC in a project, import the Python SDK in your graph code and begin registering tools or calling existing ones as demonstrated in the examples in the README.

Notes:

• The ATaC server is Python-based and uses the atac package from PyPI.
• Use uv tool to install ATaC, which aligns with the Quick Start example (uv tool install atac).
• When configuring environments, you can expose environment variables for tool paths, API endpoints, or authentication as needed by your registered MCP tools. Use an env map in your mcp_config as needed.
• If you run into issues with tool registration or tool_call failures, ensure the tool names match exactly between registration and invocation, and check that the registered tools are loaded before graph execution.
• The npm_package field is null since this is a Python MCP server, not a Node.js package.