Trail of Bits Skills MCP — ユースケース, インストール & ライブデモ

Name: Trail of Bits Skills MCP Server
Author: trailofbits

なぜ使うのか

主な機能

Written by a security firm, not a hobbyist — patterns reflect real incidents
Covers C/C++, Rust, Solidity, Python, Go, TypeScript
Defensive-only: audit/review flows, not offensive tooling
Modular — pull individual skills without adopting the whole bundle
CC-BY-SA-4.0 licensed, attribution required

ライブデモ

実際の動作

準備完了

インストール

クライアントを選択

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

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

Claude Desktop → Settings → Developer → Edit Config を開く。保存後、アプリを再起動。

~/.cursor/mcp.json · .cursor/mcp.json

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

Cursor は Claude Desktop と同じ mcpServers スキーマを使用。プロジェクト設定はグローバルより優先。

VS Code → Cline → MCP Servers → Edit

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

Cline サイドバーの MCP Servers アイコンをクリックし、"Edit Configuration" を選択。

~/.codeium/windsurf/mcp_config.json

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

Claude Desktop と同じ形式。Windsurf を再起動して反映。

~/.continue/config.json

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

Continue はマップではなくサーバーオブジェクトの配列を使用。

~/.config/zed/settings.json

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

context_servers に追加。保存時に Zed がホットリロード。

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

ワンライナー。claude mcp list で確認、claude mcp remove で削除。

ユースケース

実用的な使い方： Trail of Bits Skills

Add a security-review pass to every PR

👤 Teams without a dedicated security engineer ⏱ ~30 min intermediate

使うタイミング： Your PRs currently go straight from code-review to merge — no security eyes.

前提条件

Skills installed — git clone https://github.com/trailofbits/skills ~/.claude/skills/trailofbits

フロー

Scope the diff

Use trailofbits/pr-review skill on the current PR. List files the skill thinks are security-relevant with reasons.✓ コピーしました

→ Concrete file-by-file relevance reasons
Deep review

For each flagged file, run the appropriate domain skill (crypto, input-validation, auth). Report findings with severity.✓ コピーしました

→ Severity-ranked list with code refs
Summary for reviewer

Write a 5-bullet PR comment summarizing the findings, non-scary but precise.✓ コピーしました

→ Reviewer-friendly summary with exact line citations

結果： A security review comment on every PR that catches real issues without drowning you in false positives.

注意点

Skill flags style issues as security — Calibrate by prompting: "focus on exploitable issues only"

組み合わせ： github · filesystem

Audit cryptographic code paths in a service

👤 Devs maintaining auth/encryption code ⏱ ~90 min advanced

使うタイミング： You have a JWT/HMAC/encryption layer and want someone smart to double-check.

フロー

Find crypto

Use trailofbits/crypto-review. Scan /src for files that touch crypto APIs. Report them.✓ コピーしました

→ Crypto call-site inventory
Audit each

For each site: what algorithm? Key management? Nonce reuse risk? IV handling? Cite source.✓ コピーしました

→ Per-site audit with code refs
Remediation plan

For any medium+ finding, draft a remediation — concrete code change, not "consider using ...".✓ コピーしました

→ Applicable diffs, not vague advice

結果： A crypto audit report with concrete fixes, from a baseline of common mistakes ToB has seen in real engagements.

注意点

Not a substitute for a real audit — Use for self-review; hire professionals before launch of high-value surfaces

組み合わせ： github

組み合わせ

他のMCPと組み合わせて10倍の力を

trailofbits-skill + github

Post findings as a PR review comment

Post the review as a PR comment — severities as labels.✓ コピーしました

trailofbits-skill + filesystem

Scan the whole repo offline

Run the full security sweep across /src and write report to /audit/report.md.✓ コピーしました

ツール

このMCPが提供する機能

ツール	入力	呼び出すタイミング
scope_review	diff\|path	Before deep dive
crypto_review	path	Crypto code specifically
input_validation_review	path	API/boundary code
iac_review	path	Terraform / CloudFormation / K8s manifests

コストと制限

運用コスト

APIクォータ: N/A
呼び出しあたりのトークン: Review is reading-heavy; budget proportional to diff size
金額: Free (CC-BY-SA)
ヒント: Scope first — don't run deep skills on irrelevant files

セキュリティ

権限、シークレット、影響範囲

最小スコープ： filesystem-read

認証情報の保管： None

データ送信先： None from skill; reviewed code is in your LLM context

絶対に付与しない： write access unless auto-remediating

Not a replacement for a formal security audit — use it to catch the obvious before paying for experts

トラブルシューティング

よくあるエラーと対処法

Skill flags safe code as vulnerable

Ask for rationale; if weak, dismiss with reason. Calibration improves over sessions.

Misses obvious bugs

The skill codifies Trail of Bits' patterns — if your stack is unusual (e.g. niche Rust crate), it may miss. Add custom patterns.

Outputs too long to review

Constrain: "top 5 highest-severity findings only"

代替案

Trail of Bits Skills 他との比較

代替案	代わりに使う場面	トレードオフ
Semgrep rules	You want static analysis in CI, not interactive review	Narrower pattern set; no LLM reasoning
CodeQL	You want deep dataflow analysis	Much more setup; not a skill

その他

リソース

📖 GitHub の公式 README を読む

🐙 オープンな issue を見る

🔍 400以上のMCPサーバーとSkillsを見る