paperbanana-skill MCP — 사용 사례, 설치 & 라이브 데모

왜 쓰나요

핵심 기능

Publication-quality default style
Architecture / system diagrams
Training curve + benchmark plots
Outputs SVG + PDF (LaTeX-ready)
Consistent visual grammar across a paper's figures

라이브 데모

실제 사용 모습

paperbanana-skill.replay ▶ 준비됨

0/0

설치

클라이언트 선택

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

{
  "mcpServers": {
    "paperbanana-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/PlutoLei/paperbanana-skill",
        "~/.claude/skills/paperbanana-skill"
      ],
      "_inferred": true
    }
  }
}

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

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "paperbanana-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/PlutoLei/paperbanana-skill",
        "~/.claude/skills/paperbanana-skill"
      ],
      "_inferred": true
    }
  }
}

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

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "paperbanana-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/PlutoLei/paperbanana-skill",
        "~/.claude/skills/paperbanana-skill"
      ],
      "_inferred": true
    }
  }
}

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

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "paperbanana-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/PlutoLei/paperbanana-skill",
        "~/.claude/skills/paperbanana-skill"
      ],
      "_inferred": true
    }
  }
}

Claude Desktop과 같은 형식. Windsurf 재시작 후 적용.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "paperbanana-skill",
      "command": "git",
      "args": [
        "clone",
        "https://github.com/PlutoLei/paperbanana-skill",
        "~/.claude/skills/paperbanana-skill"
      ]
    }
  ]
}

Continue는 맵이 아닌 서버 오브젝트 배열 사용.

~/.config/zed/settings.json

{
  "context_servers": {
    "paperbanana-skill": {
      "command": {
        "path": "git",
        "args": [
          "clone",
          "https://github.com/PlutoLei/paperbanana-skill",
          "~/.claude/skills/paperbanana-skill"
        ]
      }
    }
  }
}

context_servers에 추가. 저장 시 Zed가 핫 리로드.

claude mcp add paperbanana-skill -- git clone https://github.com/PlutoLei/paperbanana-skill ~/.claude/skills/paperbanana-skill

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

사용 사례

실전 활용법: paperbanana-skill

Draw your model architecture for a conference paper

👤 ML researchers, grad students mid-paper ⏱ ~30 min intermediate

언제 쓸까: Deadline in 5 days, you still don't have a clean architecture figure.

사전 조건

Skill installed — git clone https://github.com/PlutoLei/paperbanana-skill ~/.claude/skills/paperbanana-skill

흐름

Describe the model

Use paperbanana-skill. Architecture figure: input → tokenizer → transformer stack (6 layers) → classifier head. Use the skill's default paper style.✓ 복사됨

→ SVG with clean boxes, arrows, consistent typography
Tweak

Label the transformer block internals — attention + FFN.✓ 복사됨

→ Updated figure with sub-block detail
Export for LaTeX

Export as PDF sized for a two-column paper.✓ 복사됨

→ PDF with correct aspect ratio

결과: A reviewer-worthy figure without a week of Inkscape.

함정

Figure over-decorated — Prefer restraint; one accent color, black arrows, lots of whitespace

함께 쓰기: filesystem

Turn training logs into a publication-ready plot

👤 ML researchers with wandb/tensorboard exports ⏱ ~20 min beginner

언제 쓸까: You have CSVs of training metrics and need a figure for the paper.

흐름

Feed CSV

Use paperbanana-skill. Plot loss and accuracy curves from training.csv in two-panel layout.✓ 복사됨

→ Two-panel figure with clean axes
Add baselines

Overlay the baseline model's curves (baseline.csv) as dashed.✓ 복사됨

→ Overlay added with clear legend

결과: A plot that fits the paper's aesthetic.

Keep all figures in a paper visually consistent

👤 Authors with 6+ figures drifting in style ⏱ ~45 min intermediate

언제 쓸까: You look at the paper and realize each figure has a different font.

흐름

Define the style

Use paperbanana-skill. Create a shared style: 10pt sans-serif, #1f77b4 primary, black stroke 0.8pt.✓ 복사됨

→ Style file saved
Apply to all figures

Regenerate figures 1-6 using this style.✓ 복사됨

→ Consistent batch output

결과: Figures that look like they belong in the same paper.

함께 쓰기: filesystem

조합

다른 MCP와 조합해 10배 효율

paperbanana-skill + filesystem

Organize figures in the paper's /figures directory

Save all figures under paper/figures/ with source.svg + out.pdf per figure.✓ 복사됨

paperbanana-skill + mermaid-skill

Use Mermaid for casual diagrams, PaperBanana for publication quality

Sketch in Mermaid for discussion; produce with PaperBanana for camera-ready.✓ 복사됨

도구

이 MCP가 노출하는 것

도구	입력	언제 호출
architecture_diagram	description	System figures
plot_curves	data file, metrics	Training/eval metrics
apply_style	style file, figures	Cross-figure consistency

비용 및 제한

운영 비용

API 쿼터: None
호출당 토큰: Moderate
금액: Free
팁: Lock the style once; apply across figures to avoid re-deciding.

보안

권한, 시크릿, 파급범위

자격 증명 저장: None

데이터 외부 송신: None from skill

문제 해결

자주 발생하는 오류와 해결

Exported PDF has wrong font in LaTeX

Embed fonts or use standard PDF font families; check paperbanana's export settings

SVG renders differently in browser vs LaTeX

Stick to PDF for LaTeX; SVG for web/slides

대안

paperbanana-skill 다른 것과 비교

대안	언제 쓰나	단점/장점
TikZ / PGFPlots	You want full LaTeX integration and are comfortable with TikZ	Steep learning curve
mermaid-skill	Casual docs, not publication	Not publication-quality aesthetic

paperbanana-skill