ai-first-toolkit MCP — 使用场景, 安装 & 实时演示

为什么要用

核心特性

AI-readiness audit — scores a repo on structure, docs, tests, agent-friendliness
Re-engineering recipes to improve AI-ergonomics of legacy code
Bootstrap skill for greenfield projects with sane defaults
Cross-agent: works in Claude Code and Codex
TypeScript-based tooling

实时演示

实际使用效果

ai-first-toolkit-skill.replay ▶ 就绪

0/0

安装

选择你的客户端

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

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

打开 Claude Desktop → Settings → Developer → Edit Config。保存后重启应用。

~/.cursor/mcp.json · .cursor/mcp.json

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

Cursor 使用与 Claude Desktop 相同的 mcpServers 格式。项目级配置优先于全局。

VS Code → Cline → MCP Servers → Edit

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

点击 Cline 侧栏中的 MCP Servers 图标，然后选 "Edit Configuration"。

~/.codeium/windsurf/mcp_config.json

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

格式与 Claude Desktop 相同。重启 Windsurf 生效。

~/.continue/config.json

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

Continue 使用服务器对象数组，而非映射。

~/.config/zed/settings.json

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

加入 context_servers。Zed 保存后热重载。

claude mcp add ai-first-toolkit-skill -- git clone https://github.com/techwolf-ai/ai-first-toolkit ~/.claude/skills/ai-first-toolkit

一行命令搞定。用 claude mcp list 验证，claude mcp remove 卸载。

使用场景

实战用法： ai-first-toolkit

How to audit a codebase for AI-agent readiness

👤 Tech leads preparing a codebase for heavy AI-assisted work ⏱ ~45 min intermediate

何时使用： Before the team commits to doing a lot of feature work through Claude or Codex.

前置条件

Skill cloned — git clone https://github.com/techwolf-ai/ai-first-toolkit ~/.claude/skills/ai-first-toolkit

步骤

Run the audit

Audit this repo for AI-agent readiness. Score each dimension and list the top 5 friction points.✓ 已复制

→ Dimension scores + specific files to fix
Prioritize fixes

Which of those fixes would give the biggest agent-effectiveness lift per hour of work?✓ 已复制

→ Ranked list with effort/impact
Apply the top fix

Apply fix #1 end-to-end — edit files, update docs, add tests.✓ 已复制

→ Diff + confirmation

结果： A concrete improvement plan and at least one applied fix making the repo friendlier to AI agents.

注意事项

Audit turns into a generic 'improve docs' blob — Ask for file:line level findings, not abstract advice

搭配使用： filesystem · github

Bootstrap a new AI-friendly project

👤 Founders or ICs starting a new repo ⏱ ~30 min beginner

何时使用： Day zero of a new project, before you pick bad defaults that will haunt you.

步骤

State the project

Bootstrap a new TypeScript API project using ai-first-toolkit defaults. Purpose: internal billing API.✓ 已复制

→ Proposed folder structure, tsconfig, scripts, README skeleton
Scaffold the files

Create the files. Include AGENTS.md and CLAUDE.md with the relevant codebase notes.✓ 已复制

→ Repo files materialized
First commit

Initialize git and make the first commit with a clean message.✓ 已复制

→ Clean initial repo

结果： A new repo that's friendly to both humans and AI agents from commit 0.

注意事项

Scaffolding imports dependencies you don't want — Ask Claude to list dependencies before installing — accept each one

搭配使用： github · filesystem

Re-engineer a messy module for AI readability

👤 Engineers paying down tech debt one module at a time ⏱ ~60 min advanced

何时使用： You've isolated a module that Claude keeps getting wrong — usually a sign its structure isn't agent-legible.

步骤

Diagnose the module

Look at src/billing/. What about this code makes it hard for an agent to edit safely?✓ 已复制

→ Specific red flags: indirection, hidden state, long functions
Plan the refactor

Propose a 3-step refactor that improves agent-legibility without changing public API.✓ 已复制

→ Stepwise plan with diff sketches
Execute step 1

Apply step 1. Keep tests green.✓ 已复制

→ PR-sized diff

结果： A module that Claude can actually reason about on the next iteration.

注意事项

Over-refactor and break behaviors you didn't mean to — Insist on public API stability and run tests after each step

搭配使用： filesystem · github

组合

与其他 MCP 搭配，撬动十倍杠杆

ai-first-toolkit-skill + filesystem

Audit scans the repo; filesystem is needed to actually read it

Walk the repo and score AI-readiness; report findings file-by-file.✓ 已复制

ai-first-toolkit-skill + github

Open a PR with the proposed AGENTS.md / CLAUDE.md additions

Apply the audit's top 3 fixes and open a PR titled 'AI-first improvements'.✓ 已复制

工具

此 MCP 暴露的能力

工具	输入参数	何时调用
ai-readiness-audit	repo root	Pre-investment in AI-heavy workflows
ai-first-bootstrap	project spec (lang, purpose, stack)	Greenfield projects
re-engineer-module	module path + goals	When a module resists agent edits

成本与限制

运行它的成本

API 配额: none beyond your LLM provider
每次调用 Token 数: 5–20k for an audit of a medium repo
费用: free, open source
提示: Scope audits to a subtree rather than the whole monorepo

安全

权限、密钥、影响范围

凭据存储： none

数据出站： none beyond what filesystem/github tools do

故障排查

常见错误与修复

Audit is too abstract

Re-prompt with a concrete dimension: 'evaluate just the test layout' or 'just the module boundaries'.

Bootstrap picks wrong language

Always state language+runtime+build tool in the first prompt.

替代方案

ai-first-toolkit 对比其他方案

替代方案	何时用它替代	权衡
init (built-in)	You just need a CLAUDE.md and don't care about the rest	Narrower scope
antivibe-skill	You want the opposite lens — critique of AI-shaped sludge	Diagnostic, not prescriptive

ai-first-toolkit

为什么要用

核心特性

实时演示

实际使用效果

安装

选择你的客户端

使用场景

实战用法： ai-first-toolkit

How to audit a codebase for AI-agent readiness

前置条件

步骤

注意事项

Bootstrap a new AI-friendly project

步骤

注意事项

Re-engineer a messy module for AI readability

步骤

注意事项

组合

与其他 MCP 搭配，撬动十倍杠杆

工具

此 MCP 暴露的能力

成本与限制

运行它的成本

安全

权限、密钥、影响范围

故障排查

常见错误与修复

替代方案

ai-first-toolkit 对比其他方案

更多

资源