HarnessCommunity★ 1k

MAS Sequential Thinking

Updated Jul 7, 2026

An advanced Sequential Thinking MCP server built on the Agno multi-agent framework. Every request runs through six specialized agents in parallel — Factual, Emotional, Critical, Optimistic, Creative, and Synthesis — then a Synthesis agent integrates all perspectives. Requires Python 3.10+ and an LLM API key (DeepSeek default). Optional Exa API key enables web research for four of the six agents.

View on GitHub

What it does

npx -y @smithery/cli install @FradSer/mcp-server-mas-sequential-thinking --client claude
Install the MAS Sequential Thinking MCP server for Claude Code via the Smithery CLI.
sequentialthinking
The single MCP tool exposed: routes a thought through a 6-agent parallel panel and returns synthesized analysis plus a should_continue signal for loop control.
isRevision: true
sequentialthinking parameter that marks a step as a revision of an earlier conclusion, allowing the agent panel to revisit and correct prior reasoning.

Files (1)

OVERVIEW.mdprimary · markdown · 2.2 KB

## What it does

MAS Sequential Thinking replaces the standard single-chain sequential thinking MCP tool with a 6-agent panel built on the Agno framework. Every request runs the full_exploration strategy — a fixed, deterministic multi-step path — regardless of problem complexity:

1. Initial Synthesis (Synthesis agent)
2. Parallel specialist agents (Factual, Emotional, Critical, Optimistic, Creative)
3. Final Synthesis (Synthesis agent integrates all perspectives)

The six agents and their roles:
- Factual — objective facts, data verification, web research (ExaTools optional)
- Emotional — gut reactions and intuitive pattern recognition (30s quick-reaction mode)
- Critical — risk assessment, devil's advocate, counterexample research (ExaTools optional)
- Optimistic — benefits, opportunities, success story research (ExaTools optional)
- Creative — lateral thinking, cross-industry innovation research (ExaTools optional)
- Synthesis — integrates all perspectives into a coherent, actionable response

The Synthesis agent uses an enhanced model; the five specialist agents use a standard model. Four of the six agents can perform web research via optional Exa integration.

Warning: token usage is 5-10x higher than single-agent approaches due to parallel processing.

## Install

Via Smithery CLI (recommended):
npx -y @smithery/cli install @FradSer/mcp-server-mas-sequential-thinking --client claude

Requires Python 3.10+ and an LLM API key. Set in MCP config env:
LLM_PROVIDER: deepseek (default), groq, openrouter, github, anthropic, or ollama
DEEPSEEK_API_KEY (or equivalent for your chosen provider)
EXA_API_KEY (optional, enables web research for 4 agents)

## The sequentialthinking tool

The single MCP tool exposed is sequentialthinking. Parameters include thought (the current reasoning step), thoughtNumber (1-based index), totalThoughts, nextThoughtNeeded, and isRevision.

The tool returns structuredContent.should_continue — keep calling sequentialthinking in a loop until this is false.

## When to use

Use for complex architectural decisions, ambiguous design trade-offs, or any problem that benefits from multiple cognitive perspectives simultaneously. The token cost is significant; avoid for routine tasks.