Skip to content

🛠 Tools & Actions

🛠 Tools & Actions

Tools are the building blocks of MCPStack pipelines. Each tool:

  • Encapsulates logic (e.g. querying MIMIC-IV, running Jupyter, calling scikit-longitudinal classifiers)
  • Exposes actions — Python callables that the MCP host can invoke

What is an Action?

An action is a function exposed to the LLM/host. For example, in the Hello_World tool:

from MCPStack.core.tool.base import BaseTool

class Hello_World(BaseTool):
    def actions(self):
        return [self.say_hello]

    def say_hello(self, name: str = "world") -> str:
        return f"Hello, {name}!"

MCPStack registers these actions so they can be called through the MCP server.

Lifecycle hooks

Tools can also manage backends (like databases or API clients):

  • initialize() → open connections or caches
  • teardown() → clean up
  • post_load() → called after deserialization

This ensures consistent behavior whether you load tools from JSON or build them in Python.

Example: Jupyter Tool

A Jupyter tool may expose actions like:

  • create_notebook()
  • run_notebook(path)
  • list_kernels()

Each is just a Python function returning JSON-serializable output.