Autonomous Analyst

A local MCP-powered pipeline that analyzes tabular data, detects anomalies, summarizes insights with a local LLM, and logs results for memory recall.

python

0

GitHub Stars

python

Language

7 months ago

First Indexed

3 months ago

Catalog Refreshed

Documentation & install

Readme and setup notes from the catalogue, plus a client-ready config you can copy for your MCP host.

Installation

Add the following to your MCP client configuration file.

Configuration

View docs

{
  "mcpServers": {
    "madmando-mcp-autonomous-analyst": {
      "command": "mcp",
      "args": [
        "run",
        "server.py",
        "--transport",
        "streamable-http"
      ]
    }
  }
}

Autonomous Analyst is a local, agentic AI pipeline that analyzes tabular data, detects anomalies with Mahalanobis distance, generates interpretive summaries using a local LLM, logs results to a vector store, and is fully orchestrated via the Model Context Protocol (MCP). It provides an interactive web dashboard and an automated planning tool to guide analysis steps, making it easy to explore datasets, surface insights, and recall past results.

How to use

Set up the MCP server and run the web dashboard to work with your datasets. Launch the MCP server to enable tool orchestration and autonomous planning, then start the web UI for interactive analysis.

How to install

Prerequisites: you need Python and a compatible shell environment. You also need an MCP-compatible runtime to execute the server.

Clone the project repository.
Create and activate the Python environment.
Install dependencies and run the MCP server.

Additional sections

MCP Server configuration is provided as an executable command that starts the server with a streamable HTTP transport. The following command is available to initiate the MCP server execution.

`json
{
  "mcpServers": {
    "auton_analyst": {
      "command": "mcp",
      "args": ["run", "server.py", "--transport", "streamable-http"]
    }
  }
}
`

Notes on usage and security

Use the MCP server to orchestrate all analysis steps. The pipeline includes data generation or ingestion, outlier detection via Mahalanobis distance, visualization of results, and generation of summaries using the local LLM. Results are stored in a vector store for memory recall and can be retrieved with a search utility.

Access the web dashboard at the designated port after starting the UI server. Ensure your environment variables and secrets (if any) are secured and not exposed in public environments.

Troubleshooting

If the MCP server fails to start, verify that the command path and Python environment are correct. Check that port bindings for the dashboard are available and that the vector store (ChromaDB) is accessible. Review logs for any missing dependencies or permission issues.

Available tools

generate_data

Create synthetic tabular data with Gaussian and categorical features to test the pipeline.

analyze_outliers

Label rows by computing Mahalanobis distance to identify outliers in the feature space.

plot_results

Save a visual plot that shows inliers versus outliers for quick quality assessment.

summarize_results

Interpret and explain the outlier distribution using the local LLM (llama3.2:1b).

summarize_data_stats

Describe dataset trends and statistics using the local LLM (llama3.2:1b).

log_results_to_vector_store

Store summaries in the ChromaDB vector store for persistent memory and retrieval.

search_logs

Retrieve relevant past sessions via vector search, optionally leveraging the LLM for interpretation.

autonomous_plan

Run the full pipeline and let the LLM recommend next actions automatically based on dataset context.