MCP-Server com CoConuT (Continuous Chain of Thought)

What is MCP-Server com CoConuT (Continuous Chain of Thought)?

An MCP server that provides the CoConuT tool for structured continuous chain-of-thought reasoning, with automatic cycle detection, branch management, and guided interaction. It is designed for developers using the Model Context Protocol to enhance models’ reasoning capabilities.

How to use MCP-Server com CoConuT (Continuous Chain of Thought)?

Clone the repository, install dependencies (npm install), configure settings in src/config.ts, then run with npm run dev for development or npm run build && npm start for production. The server exposes three tool variants: CoConuT (JSON), CoConuT‑MD (Markdown), and CoConuT‑HTML.

Key features of MCP-Server com CoConuT (Continuous Chain of Thought)

• Continuous chain‑of‑thought for structured problem solving
• Automatic cycle detection using Levenshtein, Jaccard, or Cosine similarity
• Branch management: explore, compare, and merge reasoning paths
• Periodic reflection every reflectionInterval thoughts
• Built‑in persistence of all reasoning data
• Multiple output formats: JSON, Markdown, HTML
• Modular architecture with dependency injection

Use cases of MCP-Server com CoConuT (Continuous Chain of Thought)

• Structuring long‑term reasoning tasks with continuous chain‑of‑thought
• Detecting and avoiding circular reasoning in model outputs
• Exploring multiple solution branches and comparing them
• Automatically reflecting on progress during problem‑solving

FAQ from MCP-Server com CoConuT (Continuous Chain of Thought)

What is the CoConuT tool?

CoConuT (Continuous Chain of Thought) is a tool that facilitates structured, step‑by‑step reasoning with automatic cycle detection, branch management, and guided user interaction.

What are the runtime dependencies?

Node.js 18 or higher and NPM are required.

How does cycle detection work?

The system uses similarity algorithms (Levenshtein, Jaccard, Cosine) with a configurable threshold (default 0.8) to detect cyclic reasoning.

Where are data persisted?

All data is automatically saved to a coconut-data folder inside the directory specified by the projectPath parameter.

What output formats are supported?

Three tool variants return results in JSON (CoConuT), Markdown (CoConuT‑MD), or structured HTML (CoConuT‑HTML).

Can I change the similarity algorithm?

Yes. The similarityAlgorithm configuration option accepts 'levenshtein' (default), 'jaccard', or 'cosine'.