Claude代码插件 是 AI Skill Hub 本期精选MCP工具之一。综合评分 8.0 分,整体质量较高。我们强烈推荐将其纳入你的 AI 工具库,帮助提升工作效率。
Claude代码插件 是一款遵循 MCP(Model Context Protocol)标准协议的 AI 工具扩展。通过 MCP 协议,它可以让 Claude、Cursor 等主流 AI 客户端直接访问和操作外部工具、数据源和服务,实现 AI 能力的无缝扩展。无论是文件操作、数据库查询还是 API 调用,都可以通过自然语言在 AI 对话中直接触发,极大提升生产效率。
Claude代码插件 是一款遵循 MCP(Model Context Protocol)标准协议的 AI 工具扩展。通过 MCP 协议,它可以让 Claude、Cursor 等主流 AI 客户端直接访问和操作外部工具、数据源和服务,实现 AI 能力的无缝扩展。无论是文件操作、数据库查询还是 API 调用,都可以通过自然语言在 AI 对话中直接触发,极大提升生产效率。
# 方式一:通过 Claude Code CLI 一键安装
claude skill install https://github.com/LeeJuOh/claude-code-zero
# 方式二:手动配置 claude_desktop_config.json
{
"mcpServers": {
"claude----": {
"command": "npx",
"args": ["-y", "claude-code-zero"]
}
}
}
# 配置文件位置
# macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
# Windows: %APPDATA%/Claude/claude_desktop_config.json
# 安装后在 Claude 对话中直接使用 # 示例: 用户: 请帮我用 Claude代码插件 执行以下任务... Claude: [自动调用 Claude代码插件 MCP 工具处理请求] # 查看可用工具列表 # 在 Claude 中输入:"列出所有可用的 MCP 工具"
// claude_desktop_config.json 配置示例
{
"mcpServers": {
"claude____": {
"command": "npx",
"args": ["-y", "claude-code-zero"],
"env": {
// "API_KEY": "your-api-key-here"
}
}
}
}
// 保存后重启 Claude Desktop 生效
A plugin marketplace I built from what I needed while using Claude Code.
<video src="https://github.com/user-attachments/assets/abb70886-6f82-474c-a956-3c89b77c4ae5" width="600" controls></video>
</div>
/plugin install <plugin-name>@claude-code-zero
Or from the terminal:
shell claude plugin add <plugin-name>@claude-code-zero ```
Full install: skills, hooks, agents, MCP servers, scripts — everything the plugin ships.
```shell
Installs SKILL.md files via the skills CLI. Quick, no marketplace registration needed — but hooks, agents, MCP servers, and scripts are not included. Plugins that depend on hooks (worktree-plus, rubber-duck-tutor, claw-mo) or agents (vision-powers, codex-advisor) will have reduced functionality.
npx skills add LeeJuOh/claude-code-zero
Problem: Codex outputs look confident but hallucinate citations and miss edge cases. You won't catch it by reading alone.
Solution: Claude independently re-verifies every Codex response — without seeing source code until after Codex returns. Five-way classification (Agreed / Disputed / Nuanced / False Positive / Uncited) so you know exactly what to trust.
codex-review · codex-adversarial · codex-rescue · codex-verify · codex-research · codex-status · codex-result · codex-cancel · codex-setup
---
Problem: Complex analysis buried in terminal text — architecture, security issues, git diffs all lose structure. Impossible to share with teammates.
Solution: Generates interactive HTML reports with Mermaid diagrams and Chart.js dashboards. Analyzes plugins (local, installed, or GitHub), git diffs, and markdown docs. Visual self-audit renders reports to PNG for verification.
plugin-visual · diff-visual · doc-visual · fact-check · context-health-visual · report-manager
---
Problem: Building a skill that triggers when it should and actually helps is trial and error — no way to measure trigger accuracy, no structured improvement loop, no benchmark to tell you if a change helped.
Solution: Anthropic's official skill-creator coaching loop — draft, test against real prompts, review side-by-side baseline-vs-with-skill diffs, improve — plus a description-trigger optimizer and an HTML benchmark viewer. auto-optimize adds a hands-off hill-climbing pass for skills that already mostly work.
skill-creator-pro · auto-optimize
---
Problem: Claude Code's built-in worktree breaks on gitignored files (.env, node_modules/), doesn't track state, and can delete uncommitted work on removal.
Solution: .worktreeinclude / .worktreelink for selective file copy and symlink. Safety guard blocks removal if uncommitted changes or unpushed commits exist. Audit trail in .worktree.log.
worktree-config · auto-hooks on WorktreeCreate/Remove
---
Problem: Your research lives in NotebookLM but querying it means context-switching, copy-pasting, and burning tokens on ungrounded answers.
Solution: Chrome automation queries NotebookLM directly from Claude Code. Source-grounded answers with automatic follow-up rounds (default 3) to fill coverage gaps. Per-project notebook registry.
notebooklm-manager (query · add · list · search · enable/disable · remove)
---
Problem: mo markdown viewer is powerful but tedious to configure — port numbers, watch patterns, fsnotify drops files silently.
Solution: Per-project config with auto-sync hook. Every time Claude writes or edits a markdown file, it appears in mo automatically. Group-based organization, full-text search, Mermaid + KaTeX + Shiki rendering.
claw-mo-setup · claw-mo-up · claw-mo-down · claw-mo-open · claw-mo-manage
---
Problem: Claude Code is trapped in a single terminal pane. Can't send commands to other panes, read their output, or orchestrate parallel workflows.
Solution: Full cmux integration — split layouts, send commands to any pane, read screen output, automate WKWebView browsers, and report progress via sidebar primitives (status / progress bar / leveled logs).
claw-mux · cmux-browser · cmux-markdown
---
Problem: WebFetch gets blocked by bot detection. Session context vanishes between conversations. Secrets end up hardcoded. References drift out of sync.
Solution: Five focused utilities — Gemini fallback (auto-triggers on WebFetch failure), resumption-ordered handoff docs, secret extraction with auto-load hooks, sitemap discovery, and reference sync.
gemini-fetch · handoff · secret-setup · fetch-sitemap · sync-references
---
Problem: Passively accepting AI-generated code leads to 17% worse comprehension. You merge code you don't truly understand.
Solution: Rubber duck questioning across 5 development phases — asks you to explain what the code does before you ship it. Hint ladder guides without revealing answers. Auto-suggests review at plan creation, PR, and git push.
duck · duck-design · duck-plan · duck-verify · duck-review · duck-orient
---
Problem: Wiring VibeProxy manually (OAuth, aliases, config.yaml, shell edits) is error-prone. State is unclear — which backend is actually responding?
Solution: Explicit state management with backup/rollback. Per-backend isolation probe tells you exactly which model handles each alias. Pre-existing alias migration with Keep / Merge / Reset modes.
setup-aliases · cc-list
---
Experimental. May require specific environments or have limited stability.
Problem: E2E tests require brittle selectors and page objects. They break on every UI change and take forever to write.
Solution: Write tests in natural language JSON. Claude reads the page, decides what to click, and validates outcomes. Video recording per test, visual regression via pixel-diff, and auto-detects your dev server (Next / Vite / Remix / Astro / and more).
e2e-test
---
/plugin disable <plugin-name>@claude-code-zero # Disable
/plugin enable <plugin-name>@claude-code-zero # Re-enable
/plugin update <plugin-name>@claude-code-zero # Update to latest
/plugin uninstall <plugin-name>@claude-code-zero # Uninstall
高质量的MCP工具,提供丰富的插件和开发功能
AI Skill Hub 为第三方内容聚合平台,本页面信息基于公开数据整理,不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后,再部署至生产环境,并做好必要的安全评估。
✅ MIT 协议 — 最宽松的开源协议之一,可自由商用、修改、分发,仅需保留版权声明。
经综合评估,Claude代码插件 在MCP工具赛道中表现稳健,质量优秀。如果你已有明确的使用需求,可以直接上手体验;如果还在评估阶段,建议对比同类工具后再做决策。
| 原始名称 | claude-code-zero |
| Topics | mcpai-agentsanthropicclaudedeveloper-tools |
| GitHub | https://github.com/LeeJuOh/claude-code-zero |
| License | MIT |
| 语言 | Python |
收录时间:2026-06-08 · 更新时间:2026-06-08 · License:MIT · AI Skill Hub 不对第三方内容的准确性作法律背书。
选择 Agent 类型,复制安装指令后粘贴到对应客户端