Skip to content

Collaborative Sense making with AI (MCP Server)

Jump to bottom
Bell Eapen edited this page Nov 10, 2025 · 1 revision

Collaborative Sense-making with AI (MCP Server)

Overview

CRISP-T provides an MCP (Model Context Protocol) server that exposes its analytical and corpus management functions as tools, resources, and prompts. This enables AI-assisted workflows where researchers can request analyses, retrieve structured outputs, and co-construct theoretical memos with an AI assistant while maintaining methodological rigor.

Capabilities

Type Examples Purpose
Tools decision_tree_classification, regression_analysis, kmeans_clustering, lstm_text_classification, semantic_chunk_search, find_similar_documents, export_metadata_df Run analyses and retrieve actionable outputs
Resources corpus://document/{id} Direct access to document text for coding
Prompts analysis_workflow, triangulation_guide Guided step-by-step procedures aligning with INSTRUCTION.md

Sense-making Workflow with AI

  1. Load corpus and request --topics and --assign via AI-guided CLI or direct tool invocation.
  2. Ask the assistant to run regression_analysis or decision_tree_classification and summarize key predictors.
  3. Use semantic_chunk_search to retrieve exemplar passages supporting or challenging numeric findings.
  4. Co-author analytic memos: the assistant drafts, the researcher validates and refines.
  5. Visualize with crispviz and have the assistant annotate figure insights for triangulation.

Example Interactions

  • "Find documents similar to DOC101 and summarize their common themes."
  • "Run logistic regression with high_fatigue as outcome; report odds ratios for sleep_balance and inflammation_marker."
  • "Search for chunks related to 'policy clarity' in DOC450 with similarity > 8.8 and compile exemplars."

Best Practices

  • Treat AI outputs as memos-in-progress; maintain auditability by saving tool outputs.
  • Use high thresholds for confirmatory chunk search; lower thresholds for exploratory sampling.
  • Regularly link assistant-generated summaries to concrete outputs (coefficients, feature importances).

Limitations

  • AI-generated text can drift; anchor to repository outputs and explicit corpus metadata.
  • Some advanced validations (e.g., silhouette scores) may not be available—document as future work.

References

  • CRISP-T README (MCP section)
  • notes/DEMO.md and notes/INSTRUCTION.md for usage patterns

See Also

Clone this wiki locally