Skip to content

🚀 Getting Started

🚀 Getting Started

This quickstart walks you through:

  1. Installing MCPStack
  2. Building a pipeline with the hello_world tool — both via CLI and programmatically
  3. Running it with an MCP host configuration (fastmcp or Claude Desktop)

1. Installation

We recommend installing via uv, which provides fast, isolated environments. If you prefer pip, it works the same.

uv add mcpstack

python -m venv .venv
source .venv/bin/activate
pip install mcpstack

2. Creating a pipeline with Hello_World

The hello_world tool is bundled with MCPStack. It exposes a single action (say_hello) to demonstrate the tool interface.

mcpstack pipeline hello_world --new-pipeline my_pipeline.json
This command creates a new pipeline JSON with the hello_world tool inside. To inspect your pipeline: 
cat my_pipeline.json

You can also build pipelines directly in Python: 

from MCPStack.stack import MCPStackCore
from MCPStack.tools.hello_world.hello_world import Hello_World

stack = MCPStackCore()
stack = stack.with_tool(Hello_World())
stack.save("my_pipeline.json")

3. Running the pipeline with a host

MCPStack can generate and run host configurations.

mcpstack run --pipeline my_pipeline.json --config-type fastmcp

MCPStack will:

  1. Load your pipeline
  2. Generate a fastmcp configuration
  3. Start the MCP server
mcpstack build --pipeline my_pipeline.json --config-type claude

Claude supports local MCP servers. You can export the config and point Claude to it. Then open Claude Desktop and enjoy your pipeline being called by the LLM.

Tip

You can always combine multiple tools into the same pipeline — e.g. mimic, jupyter, scikit-longitudinal, and more!