Bayesian

Bayesian MCP provides a server that lets you perform Bayesian inference, model comparison, and predictive analytics, enabling LLM-driven probabilistic reasoning with uncertainty quantification.

How to use

You interact with the Bayesian MCP server through an MCP client. Start a local server or connect to a remote MCP endpoint, then send requests to create models, update beliefs with new data, generate predictions from the posterior, compare different models, and visualize posterior distributions. The server exposes core functions that let you define probabilistic models, update them as new evidence arrives, and obtain uncertainty-aware predictions for future data or conditions.

How to install

Prerequisites you need before installing the Bayesian MCP server:

• Python 3.9+

• PyMC 5.0+

• ArviZ

• NumPy

• Matplotlib

• FastAPI

• Uvicorn

Installation steps (run these in a terminal):

git clone https://github.com/wrenchchatrepo/bayesian-mcp.git
cd bayesian-mcp
pip install -e .

# Optional: create a virtual environment first
# python -m venv venv
# source venv/bin/activate  (Linux/macOS)
# venv\Scripts\activate     (Windows)

Starting the server

To run the server locally with default settings, start the Python script that launches the MCP server.

python bayesian_mcp.py

Practical start variants

If you want to explicitly bind to a host and port or adjust logging, you can start with additional options.

python bayesian_mcp.py --host 0.0.0.0 --port 8080 --log-level debug

Available tools

create_model

Create a new Bayesian model with specified variables, priors, and likelihood structure.

update_beliefs

Update the model beliefs with new observed data using sampling to approximate the posterior.

predict

Generate predictions from the posterior distribution under specified conditions.

compare_models

Compare multiple models using information criteria such as WAIC or other metrics.

create_visualization

Produce visualizations of posterior distributions and related diagnostics.