Recursive Companion
Orchestrates iterative refinement of text outputs using self-critique cycles to achieve high-quality, convergent results.
About
Recursive Companion is an MCP (Model Context Protocol) server designed for advanced text generation and refinement. It implements a 'Draft → Critique → Revise → Converge' methodology, breaking down complex content creation into manageable, iterative steps. This process leverages multiple AI models, including faster models for parallel critiques, to continuously improve an initial draft until it reaches a specified convergence threshold. By providing incremental processing and progress visibility, it avoids timeouts and enhances the user experience for generating refined, high-quality content across various domains like technical, marketing, or legal.
Key Features
- Progress Visibility with immediate step returns for UI updates
- Support for multiple Parallel Sessions
- Incremental Refinement to prevent timeouts
- 8 GitHub stars
- Mathematical Convergence tracking via cosine similarity
- Domain-Specific Optimization with auto-detection and specialized prompts
Use Cases
- Generating highly refined and accurate technical documentation
- Crafting polished marketing copy or strategic plans through iterative improvements
- Creating legally precise or financially sound content by self-critique and revision cycles