Server facts
discovery-engine MCP Server
Project
- Author
- leap-laboratories
- License
- MIT
- Category
- Data & Analytics
Tech
- Runtime / language
- python
- Transports
- stdio
- Remote available
- No
- Authentication
- Not specified
Activity
- GitHub stars
- 7
- Last commit
- Data as of
Tags
- Data Analysis
- agent tool
- ai for science
- analysis
- analytics
- anomaly explanation
- automated analysis
- automated eda
- automated science
- automl alternative
- biomarker discovery
- clinical trial analysis
- combinatorial patterns
- csv analysis
- data driven
- data driven science
- data first
- data mining
- dataset analysis
- eda
- explainable ai
- exploratory data analysis
- feature importance
- feature interactions
- hold out validation
- hypothesis free
- hypothesis generation
- interaction effects
- interpretable ml
- knowledge discovery
- literature search
- mcp server
- non linear patterns
- novelty detection
- p values
- pattern finding
- pattern recognition
- research tool
- root cause analysis
- scientific discovery
- statistical validation
- subgroup analysis
- subgroup discovery
- tabular data
- xai
Overview
Disco searches combinations of feature conditions to explain changes in a chosen target column, supporting analysts and agents who have not yet formed a specific hypothesis. Its discovery pipeline learns patterns from models, tests them against held-out observations, and adjusts statistical significance for multiple comparisons. Results describe the conditions behind each finding alongside effect sizes and adjusted p-values. Literature context and novelty assessment are available with LLM processing, which public analyses always use and private analyses can enable. The focus is exploratory pattern discovery rather than routine querying or descriptive analysis.
The MCP workflow starts with discovery_estimate to assess credit costs, followed by discovery_upload and discovery_analyze to submit the dataset and target. Analysis returns a run identifier; discovery_status tracks progress, and discovery_get_results retrieves completed findings and links to report views for patterns, features, and summaries. Public analyses publish both datasets and results at no charge, while private analyses consume credits. Account tools support email-based signup and login, balance checks, plan selection, payment-method attachment, and credit purchases.
discovery-engine setup notes
- discovery-engine records stdio only; confirm a network transport upstream before planning remote access.