Neha-Rudraraju/MCP

MCP: Data Analyst Duo

A lightweight two-agent pipeline using the Model Context Protocol (MCP) to load a CSV, compute statistics & correlations, interpret results with simple rule-based insights, and assemble a Markdown report—all powered by a Gradio interface.

Features

• —ComputeAgent
• —Loads any publicly-accessible CSV into a pandas DataFrame
• —Computes descriptive statistics (count, mean, std, min, quartiles, max)
• —Computes a Pearson correlation matrix for all numeric columns

• —InterpretAgent
• —Acknowledges each compute result
• —Picks top 3 columns by range and top 3 strongest correlations
• —Formats human-readable insights and a concise summary
• —Assembles a final Markdown report combining preview, metrics, and insights

• —Gradio UI
• —Input: Paste your CSV URL
• —Outputs (in order):
• —First five rows preview
• —JSON statistics
• —JSON correlation matrix
• —ComputeAgent message history
• —InterpretAgent message history
• —Final Markdown report

Architecture

1. 1.MCPMessage
2. 2.Typed messages (request_*, *_result, ack, *_interpretation, report_result)
3. 3.Ensures strict request–response semantics

1. 1.ComputeAgent
2. 2.Handlers for request_data_load, request_statistics, request_correlation
3. 3.Emits data_load_result, statistics_result, correlation_result

1. 1.InterpretAgent
2. 2.Receives compute results, replies with ack
3. 3.Runs interpret_statistics & interpret_correlation tools
4. 4.Emits statistics_interpretation, correlation_interpretation
5. 5.On request_report, emits report_result with assembled Markdown

Usage

1. 1.Clone or fork this Space.
2. 2.Ensure your CSV is publicly accessible (e.g., hosted on GitHub, S3, etc.).
3. 3.In the Gradio UI, paste the CSV URL and click Run.
4. 4.Inspect each output panel—preview, raw metrics, histories, then read the final report summary.