Grainulator: The MCP-Powered Research Plugin That Forces Claude Code to Prove Its Claims

Grainulator is a Claude Code plugin that transforms vague research prompts into a structured, auditable process. It generates typed claims, detects conflicts, and scores confidence to force AI to provide evidence for its findings.

Details

Grainulator is a Claude Code plugin that transforms vague research prompts into a structured, auditable process. When asked to research a topic, it orchestrates a multi-pass workflow where every finding becomes a typed claim stored in a claims.json file. These claims are then adversarially challenged, confidence-graded, and compiled into decision-ready briefs. The core innovation is the use of 'evidence tiers' that grade claim confidence from 'stated' to 'web' to 'documented' to 'tested' to 'production'. The compiler runs 7 passes over the claims, checking for type coverage, evidence strength, conflict detection, and bias. If unresolved conflicts exist, it blocks output until they are resolved. Grainulator uses an MCP (Model Context Protocol) architecture with four specialized servers: 'wheat' for claims management, 'mill' for format conversion, 'silo' for persistent knowledge storage, and 'DeepWiki' for codebase research capabilities. This modular design allows each component to be updated independently and potentially used by other Claude Code plugins. The plugin is designed for use cases beyond basic code questions, such as architecture decisions, library comparisons, codebase audits, onboarding research, and documentation generation. The adversarial challenge system is particularly valuable for catching AI hallucinations by forcing Claude to provide evidence before accepting claims.