/ 디렉터리 / 플레이그라운드 / mcp-sequentialthinking-tools
● 커뮤니티 spences10 ⚡ 바로 사용

mcp-sequentialthinking-tools

제작: spences10 · spences10/mcp-sequentialthinking-tools

Force your agent to think step-by-step AND suggest which other MCP tool to invoke at each step — a meta-planner for multi-tool workflows.

A fork of the canonical Sequential Thinking MCP that goes one step further: at every thought, it scores and recommends the best MCP tools from your installed set. Useful when Claude has 10+ MCPs loaded and keeps picking the wrong one.

왜 쓰나요

핵심 기능

라이브 데모

실제 사용 모습

sequentialthinking-tools.replay ▶ 준비됨
0/0

설치

클라이언트 선택

~/Library/Application Support/Claude/claude_desktop_config.json  · Windows: %APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "sequentialthinking-tools": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-sequentialthinking-tools"
      ],
      "_inferred": true
    }
  }
}

Claude Desktop → Settings → Developer → Edit Config 열기. 저장 후 앱 재시작.

~/.cursor/mcp.json · .cursor/mcp.json
{
  "mcpServers": {
    "sequentialthinking-tools": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-sequentialthinking-tools"
      ],
      "_inferred": true
    }
  }
}

Cursor는 Claude Desktop과 동일한 mcpServers 스키마 사용. 프로젝트 설정이 전역보다 우선.

VS Code → Cline → MCP Servers → Edit
{
  "mcpServers": {
    "sequentialthinking-tools": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-sequentialthinking-tools"
      ],
      "_inferred": true
    }
  }
}

Cline 사이드바의 MCP Servers 아이콘 클릭 후 "Edit Configuration" 선택.

~/.codeium/windsurf/mcp_config.json
{
  "mcpServers": {
    "sequentialthinking-tools": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-sequentialthinking-tools"
      ],
      "_inferred": true
    }
  }
}

Claude Desktop과 같은 형식. Windsurf 재시작 후 적용.

~/.continue/config.json
{
  "mcpServers": [
    {
      "name": "sequentialthinking-tools",
      "command": "npx",
      "args": [
        "-y",
        "mcp-sequentialthinking-tools"
      ]
    }
  ]
}

Continue는 맵이 아닌 서버 오브젝트 배열 사용.

~/.config/zed/settings.json
{
  "context_servers": {
    "sequentialthinking-tools": {
      "command": {
        "path": "npx",
        "args": [
          "-y",
          "mcp-sequentialthinking-tools"
        ]
      }
    }
  }
}

context_servers에 추가. 저장 시 Zed가 핫 리로드.

claude mcp add sequentialthinking-tools -- npx -y mcp-sequentialthinking-tools

한 줄 명령. claude mcp list로 확인, claude mcp remove로 제거.

사용 사례

실전 활용법: mcp-sequentialthinking-tools

How to plan a multi-MCP workflow with sequentialthinking-tools

👤 Power users with many MCPs installed ⏱ ~5 min beginner

언제 쓸까: You have Sentry + GitHub + Linear + Playwright + Notion all loaded and Claude keeps calling the wrong one or skipping steps.

흐름
  1. Ask for a plan before execution
    Before doing anything, use sequentialthinking_tools to plan: 'Triage the top Sentry issue, find the culprit commit on GitHub, file a Linear ticket.' Give 5 steps with tool recommendations.✓ 복사됨
    → Numbered plan with tool names + confidence scores
  2. Execute one step at a time
    Execute step 1. After done, show me the result and the next-step recommendation.✓ 복사됨
    → Clean handoff between tools without skipping checks

결과: Predictable tool usage, fewer wrong-tool detours, and an audit trail of reasoning.

함정
  • Model still picks a wrong tool despite the plan — Explicitly reject the bad choice in the next turn; the planner learns in-context

Break reasoning loops with sequentialthinking-tools

👤 Developers building agents that get stuck retrying ⏱ ~10 min intermediate

언제 쓸까: Your agent is looping on the same failing approach and not re-evaluating.

흐름
  1. Force a revision
    Use sequentialthinking_tools with is_revision=true to revisit step 3 of your current plan. What assumption was wrong?✓ 복사됨
    → A named assumption and an alternative tool or approach
  2. Branch into a parallel plan
    Create a branch from step 3 and propose a different tool path. Compare confidence to the original.✓ 복사됨
    → Two parallel paths scored

결과: Agent breaks out of the loop with a reasoned pivot.

함께 쓰기: vibe-check

조합

다른 MCP와 조합해 10배 효율

sequentialthinking-tools + vibe-check

Plan + sanity-check — sequentialthinking plans the steps, vibe-check challenges the plan before execution

Plan a refactor with sequentialthinking_tools, then call vibe_check to poke holes before I run it.✓ 복사됨
sequentialthinking-tools + shrimp-task-manager

Turn each planned step into a persistent task

Plan with sequentialthinking_tools, then hand each step to shrimp-task-manager as a tracked task.✓ 복사됨

도구

이 MCP가 노출하는 것

도구입력언제 호출비용
sequentialthinking_tools thought: str, next_thought_needed: bool, thought_number: int, total_thoughts: int, is_revision?: bool, branch_from_thought?: int Start of any multi-step or multi-MCP task where tool choice matters 0 — local reasoning wrapper

비용 및 제한

운영 비용

API 쿼터
None — runs locally
호출당 토큰
200–600 per thought
금액
Free
Cap total_thoughts at 5–7; longer chains rarely add value and spend tokens

보안

권한, 시크릿, 파급범위

자격 증명 저장: None required
데이터 외부 송신: None — 100% local

문제 해결

자주 발생하는 오류와 해결

Tool recommendations are generic / unrelated to installed MCPs

The recommender works from names in the prompt — include a short list of your installed MCPs in the system prompt

Agent ignores the plan and jumps to action

Explicitly say 'do not call any other tool until you finish the sequential plan'

npx can't find the package

Try npm cache clean --force or pin a version: npx -y mcp-sequentialthinking-tools@latest

확인: npx mcp-sequentialthinking-tools --help

대안

mcp-sequentialthinking-tools 다른 것과 비교

대안언제 쓰나단점/장점
Sequential Thinking MCP (canonical)You don't need tool recommendations, just step-by-step thinkingNo tool suggestions — the main upside of this fork is lost
vibe-checkYou want metacognitive feedback instead of sequential decompositionDifferent shape — vibe-check interrupts, this one plans

더 보기

리소스

📖 GitHub에서 공식 README 읽기

🐙 열린 이슈 보기

🔍 400+ MCP 서버 및 Skills 전체 보기