freemcplab / 디렉터리 / 플레이그라운드 / Claude Octopus

● 커뮤니티 nyldn 🔑 본인 키 필요

Claude Octopus

Name: Claude Octopus (Claude Skill)
Author: nyldn

Run up to 8 models in parallel on the same task; surface where they disagree — that's where your blindspot is.

Octopus is a multi-model orchestration skill: define a task, it dispatches to N model providers, then highlights consensus and disagreement. Useful for research, design review, and plan critique where one model can have a confident-but-wrong opinion.

▶ 데모 보기 📦 설치 💡 사용 사례

왜 쓰나요

핵심 기능

Up to 8 models in parallel
Disagreement highlighting
Cost guard (caps spend per run)
Supports Anthropic, OpenAI, Google, xAI, etc.
Output is a single decision-aid Markdown

라이브 데모

실제 사용 모습

준비됨

설치

클라이언트 선택

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "claude-octopus": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/nyldn/claude-octopus"
      ],
      "_inferred": true
    }
  }
}

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

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "claude-octopus": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/nyldn/claude-octopus"
      ],
      "_inferred": true
    }
  }
}

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

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "claude-octopus": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/nyldn/claude-octopus"
      ],
      "_inferred": true
    }
  }
}

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

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "claude-octopus": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/nyldn/claude-octopus"
      ],
      "_inferred": true
    }
  }
}

Claude Desktop과 같은 형식. Windsurf 재시작 후 적용.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "claude-octopus",
      "command": "git",
      "args": [
        "clone",
        "https://github.com/nyldn/claude-octopus"
      ]
    }
  ]
}

Continue는 맵이 아닌 서버 오브젝트 배열 사용.

~/.config/zed/settings.json

{
  "context_servers": {
    "claude-octopus": {
      "command": {
        "path": "git",
        "args": [
          "clone",
          "https://github.com/nyldn/claude-octopus"
        ]
      }
    }
  }
}

context_servers에 추가. 저장 시 Zed가 핫 리로드.

claude mcp add claude-octopus -- git clone https://github.com/nyldn/claude-octopus

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

사용 사례

실전 활용법: Claude Octopus

Multi-model design review

👤 Claude Code users ⏱ ~15 min intermediate

언제 쓸까: You're about to commit to an architecture and want a second (and third, and fourth) opinion.

흐름

Brief

Provide the design doc + question.✓ 복사됨

→ Octopus dispatches to N models
Read disagreements

Skim only where models disagree.✓ 복사됨

→ Real risks surface
Decide

Resolve disagreements with a focused follow-up.✓ 복사됨

→ Documented decision

결과: Decision made with explicit awareness of where models disagree.

조합

다른 MCP와 조합해 10배 효율

claude-octopus + flow-next-skill

Use Octopus on the proposed plan, then flow-next executes the agreed-on version

Combine claude-octopus with flow-next-skill: Use Octopus on the proposed plan, then flow-next executes the agreed-on version✓ 복사됨

도구

이 MCP가 노출하는 것

도구	입력	언제 호출	비용
octopus-run	(see docs)	Dispatch to N model providers	1 call

비용 및 제한

운영 비용

API 쿼터: Per-provider limits
호출당 토큰: N × normal
금액: Pay-per-call across providers
팁: Set the cost cap; use cheap models for the panel and one premium

보안

권한, 시크릿, 파급범위

최소 스코프: multi-provider keys

자격 증명 저장: API keys per provider in env

데이터 외부 송신: Each configured provider

절대 부여 금지: ship customer data through every provider blindly — review egress

문제 해결

자주 발생하는 오류와 해결

One provider 401s

Octopus continues with the rest; log shows which key

대안

Claude Octopus 다른 것과 비교

대안	언제 쓰나	단점/장점
Single-model with self-critique	Cost-sensitive	Misses provider-specific blindspots