Vibe Preprocessing and Analysis MCP Server

# Vibe Preprocessing and Analysis MCP Server for CSV files A powerful MCP (Model Control Protocol) server for preprocessing and analyzing CSV files. This server provides a suite of tools for data manipulation, visualization, and analysis through an intuitive interface. ## Features - **Data Loading and Management** - Load CSV files from a specified working directory - Set and manage working directories - List files in the working directory - Save processed dataframes to new files - **Data Preprocessing** - Handle mixed data types in columns - Manage null values with various strategies: - Remove rows with nulls - Fill with mean/median/mode - Forward/backward fill - Fill with constant values - Drop and rename columns - Run custom dataframe editing code - Save processed data to new files - **Data Analysis** - Generate comprehensive data descriptions - Create correlation matrices with visualizations - Handle mixed data types in columns - Run custom analysis code - **Data Visualization** - Create various types of plots: - Line plots - Bar charts - Scatter plots - Histograms with KDE - Box plots - Violin plots - Pie charts - Count plots - Kernel Density Estimation plots - Custom graph generation through code - Save visualizations to the working directory - Run custom visualization code ## Setup Instructions ### Prerequisites - Python 3.x - uv (recommended package manager). I recommend using [uv](https://docs.astral.sh/uv/) to manage the server. ### Installation 1. Add MCP and required dependencies: ```bash uv add "mcp[cli]" uv add pandas matplotlib seaborn numpy ``` 2. Install the server in Claude Desktop: ```bash mcp install server.py ``` ### Alternative Installation with pip If you prefer using pip: ```bash pip install "mcp[cli]" pandas matplotlib seaborn numpy ``` ## Usage 1. Start the MCP server: ```bash uv run mcp ``` 2. Test the server using MCP Inspector: ```bash mcp dev server.py ``` You can install this server in [Claude Desktop](https://claude.ai/download) and interact with it right away by running: ```bash mcp install server.py ``` Alternatively, you can test it with the MCP Inspector: ```bash mcp dev server.py ``` ## Available Tools ### Data Management - `send_work_dir()`: Retrieve the current working directory - `set_work_dir(new_work_dir)`: Set a new working directory - `list_work_dir_files()`: List files in the current working directory - `load_csv(filename)`: Load a CSV file into the system - `save_global_df(filename)`: Save the current dataframe to a file ### Data Preprocessing - `handle_column_mixed_types()`: Handle columns with mixed data types - `handle_null_values(strategy, columns)`: Handle null values in the dataset with various strategies - `drop_columns(columns)`: Remove specified columns - `rename_columns(column_mapping)`: Rename columns in the dataframe - `run_custom_df_edit_code(code)`: Execute custom dataframe manipulation code ### Data Analysis - `describe_df()`: Generate a statistical summary of the dataframe - `generate_correlation_matrix()`: Create a correlation matrix with visualization ### Data Visualization - `plot_graph(graph_type, x_column, y_column, output_filename)`: Create various types of plots - Supported graph types: line, bar, scatter, hist, box, violin, pie, count, kde - `run_custom_graph_code(code)`: Execute custom visualization code ## Environment Variables - `WORK_DIR`: The working directory where files are read from and saved to ## Error Handling The server includes comprehensive error handling for: - Missing working directories - File not found errors - Data loading and processing errors - Invalid operations on empty dataframes - Mixed data type handling - Custom code execution errors - Invalid column names - Invalid graph types - Null value handling errors ## Contributing Feel free to submit issues and enhancement requests!