FradSer/mcp-server-mas-sequential-thinking

What it solves

This project provides a way for LLM clients (like Claude Desktop) to perform deeper, multi-perspective analysis of a problem. Instead of a single response, it implements a structured "sequential thinking" process where a problem is analyzed from multiple cognitive perspectives simultaneously to avoid narrow reasoning and ensure a more comprehensive final answer.

How it works

It operates as an MCP (Model Context Protocol) server that exposes a single tool called sequentialthinking. When this tool is called, it triggers a fixed multi-agent workflow:

1. Initial Synthesis: A synthesis agent creates an initial overview of the thought.
2. Parallel Specialist Agents: Six specialized agents (Factual, Emotional, Critical, Critical, Optimistic, Creative, and Meta-cognitive) analyze the input from their specific focus areas. These agents can optionally use web research via ExaTools.
3. Final Synthesis: A synthesis agent integrates the findings from all specialists into a unified response.

The system is designed to be used in a multi-step loop where the LLM client continues calling the tool until a final conclusion is reached, supporting revisions and branching of thoughts.

Who it’s for

Users of MCP-compatible LLM clients who need high-quality, rigorous reasoning and are willing to trade higher token consumption for deeper analysis.

Highlights

• Multi-Perspective Analysis: Uses a team of six specialist agents to cover factual, emotional, critical, and creative angles.
• Deterministic Workflow: Every request follows the same rigorous path of synthesis $\rightarrow$ parallel analysis $>\rightarrow$ final synthesis.
• Flexible Provider Support: Compatible with DeepSeek, Groq, OpenRouter, Anthropic, and Ollama.
• Optional Web Research: Integrates with ExaTools for real-time data retrieval during the specialist phase.
• Structured Loop: Provides a canonical continuation signal (should_continue) to guide the LLM client through complex reasoning steps.