[!IMPORTANT] If you have not been across the MCPStack main orchestrator repository, please start there: View MCPStack

💡 About The MCPStack MIMIC Tool

MCPStack MIMIC is an MCP tool that connects the MIMIC-IV clinical database (with either SQLite or BigQuery backends) into your MCPStack pipelines.

In layman's terms:

• MIMIC-IV is a large, de-identified database of ICU patient records, commonly used for healthcare research.
• This tool makes that dataset accessible to an LLM in a controlled way.
• It provides actions like listing available tables, showing table structure with sample data, and running queries; all exposed through MCP so your model can reason with healthcare data securely.

What is MCPStack, in layman's terms?

The Model Context Protocol (MCP) standardises how tools talk to large language models. MCPStack is the orchestrator: it lets you stack multiple MCP tools together into a pipeline and then expose them inside an LLM environment (like Claude Desktop).

Think of it like scikit-learn pipelines, but for LLMs:

• In scikit-learn: you chain preprocessors, transformers, estimators.
• In MCPStack: you chain MCP tools (like MIMIC, Jupyter Notebook MCP, etc).

[!IMPORTANT] This MCP has been made possible thanks to the M3 original work by @rafiattrach, @rajna-fani, @MoreiraP12 Under Dr. Leo Celi's supervision at MIT Lab for Computational Physiology. Following a first pull request of MCPStack to M3, we realised that we needed to externalise MCPStack to make it more modular and reusable across different use-cases. As such, MCPStack MIMIC is a copy of the original M3 codebase, with adjustments only based on how how MCPStack works, and how it is structured.

---

Installation

You can install the MIMIC tool as a standalone package. Thanks to pyproject.toml entry points, MCPStack will auto-discover it.

PyPI Installation Via UV

uv add mcpstack_mimic

PyPI Installation Via pip

pip install mcpstack-mimic

Install pre-commit hooks (optional, for development)

uv run pre-commit install
# or pip install pre-commit

Run Unit Tests (optional, for development)

uv run pytest

---

🔌 Using With MCPStack

The MIMIC tool is auto-registered in MCPStack through its entry points:

[project.entry-points."mcpstack.tools"]
mimic = "mcpstack_mimic.tools.mimic.mimic:MIMIC"

That means MCPStack will “see” it without any extra configuration.

Initialise the database

For SQLite (demo dataset by default):

uv run mcpstack tools mimic init --dataset mimic-iv-demo

This downloads and prepares the dataset locally.

Configure the tool

Pick a backend (SQLite or BigQuery):

uv run mcpstack tools mimic configure --backend sqlite --db-path ./mimic.db

or

uv run mcpstack tools mimic configure --backend bigquery --project-id <YOUR_GCP_PROJECT>

This generates a mimic_config.json you can later feed into pipelines.

Check status

uv run mcpstack tools mimic status

[!NOTE] We favourite uv for running MCPStack commands, but you can also use mcpstack directly if installed globally with pip install mcpstack.

---

🖇️ Build A Pipeline With MIMIC

Now that the tool is installed and configured, add it to your pipeline:

Default MIMIC Pipeline (Runs with demo MIMIC dataset)

uv run mcpstack pipeline mimic --new-pipeline my_pipeline.json

Create a new pipeline and add MIMIC previously custom-configured

uv run mcpstack pipeline mimic --new-pipeline my_pipeline.json --tool-config mimic_config.json

Or append to an existing pipeline

uv run mcpstack pipeline mimic --to-pipeline existing_pipeline.json --tool-config mimic_config.json

Run it inside Claude Desktop

uv run mcpstack build --pipeline my_pipeline.json

Your LLM can now use the MIMIC tool in conversation, with secure access to the clinical dataset. Open Claude Desktop, and tada!

---

📖 Programmatic API

from mcpstack_mimic.tools.mimic.mimic import MIMIC
from mcpstack_mimic.tools.mimic.backend.backends.sqlite import SQLiteBackend
from mcpstack.stack import MCPStackCore

pipeline = (
    MCPStackCore() #define =config if needed
    .with_tool(MIMIC(
        backends=[
            SQLiteBackend("<path_to_your_mimic.db>")  # SQLite backend with local MIMIC-IV database
        ])
    # Here you can add as many as new `.with_tool(.)` of interest to play with.
    ).build(
        type="fastmcp",
        save_path="my_mimic_pipeline.json",
    ).run()
)

[!IMPORTANT] The current repository has (1) technical debts, as in it would benefit from a refactor to make it maybe less messy; for instance, organising the actions into specific files. (2) a lack of documentation, the readme could deserve more in depth exploration of all the possible configurations, explore the code if you are a developer ; it is a little codebase. Pull Requests are more than welcome to minimise the tech debts and improve the documentation.

---

📽️ Video Demo

🔐 License

MIT — see LICENSE.