SkillPilot

Describe your project → get a complete AI agent team with MCP servers, curated skills, and step-by-step roadmap.

SkillPilot is an MCP (Model Context Protocol) server that turns a single project description into a full implementation plan. It works inside any IDE that supports MCP — Cursor, Claude Code, Windsurf, Codex, and more.

What it does

1. You describe your project — "veterinary clinic website with booking and payments"

2. SkillPilot returns a complete plan:

   • MCP agents to install (from 60+ curated catalog + 37K on GitHub)
   • Ready-to-use AI skills with full instructions (auth, payments, database, design, security...)
   • Open source projects to reference or fork
   • All API keys needed with direct links
   • Step-by-step roadmap for your IDE agent

3. One command installs everything — .mcp.json, .env.example, .cursorrules, CLAUDE.md, ROADMAP.md

Agent Team System — parallel swarm with 14 roles

Describe your project and get a full team of specialized AI agents:

"Use skillpilot_team for an online marketplace with payments and AI search"

SkillPilot assembles a team from 14 agent roles, each with 2 sub-agents and community skills as weapons:

| Priority | Roles | Focus | |----------|-------|-------| | 1 — Foundation | Backend Architect, Auth & Security Lead | Server, API, database, auth | | 2 — Business Logic | Frontend, Payments, AI/ML, Communications, Business Analyst, Strategy Consultant | UI, billing, AI, email, business planning | | 3 — Quality | QA Lead, Code Review Lead, Performance Engineer | Testing, review, optimization | | 4 — Launch | DevOps, Analytics, Content & CMS, Project Coordinator | Deploy, metrics, content, PM |

Each agent is armed with:

• MCP weapons — real MCP servers (Supabase, Stripe, Vercel, Firecrawl...)
• Prompt weapons — expert instructions (System Architect, Security Expert, QA Engineer...)
• Skill weapons — community skills fetched from 8 GitHub sources (2,300+ skills)

Team sizes: Squad (8 agents) → Platoon (16) → Company (24) → Battalion (28)

⚔️ Battle Mode — adversarial evaluation

Instead of one plan, generate 3 competing strategies and let your IDE's AI pick the best:

"Use skillpilot_plan in battle mode for an online marketplace"

| Team | Priority | Philosophy | |------|----------|------------| | 🏎️ SPEED | Time-to-market | Managed services, ship in days | | 🏗️ SCALE | Scalability | Open-source, self-hosted, 10x ready | | 💰 BUDGET | Minimize cost | Free tiers, near-zero running costs |

Each team proposes a full stack with agents and skills. Your AI then:

1. Scores each strategy on 6 weighted criteria
2. Identifies strengths and weaknesses
3. Picks the winner with justification
4. Asks you to confirm
5. Implements only the winning plan

NEW: Business planning agents

v0.5.0 adds Business Analyst and Strategy Consultant roles. When your project description mentions business plan, investment, ROI, or similar keywords — these agents join the team automatically.

"Use skillpilot_team for a car rental service in Batumi, Georgia with business plan"

Business Analyst builds financial models: P&L projections, NPV/IRR, break-even, unit economics, depreciation, cash flow. Strategy Consultant analyzes the market: Porter's Five Forces, PEST, SWOT, customer segmentation, go-to-market.

Both agents use Firecrawl (web scraping) to research the actual market before building models — every number must cite a source or explain the assumption.

Powered by 50 business SKILL.md files from linuszz/business-strategy-planning-skills.

How it works

SkillPilot is a local MCP server — no cloud, no hosting, no deployment needed. Your IDE launches it as a local process and communicates via stdin/stdout. All data stays on your machine.

Quick Start

Option A: Install from npm

npm install -g @exz1k/skill-pilot

Option B: Clone and build

git clone https://github.com/EXZ1K/skillpilot.git
cd skillpilot
npm install
npm run build

Connect to your IDE

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "skillpilot": {
      "command": "npx",
      "args": ["-y", "@exz1k/skill-pilot", "--mcp"],
      "env": {
        "GITHUB_TOKEN": "your_github_token"
      }
    }
  }
}

Claude Code

claude mcp add skillpilot -- npx -y @exz1k/skill-pilot --mcp

Windsurf / Codex / any MCP-compatible IDE

Same pattern — use npx -y @exz1k/skill-pilot --mcp as the command.

Use it

"Use skillpilot_plan for [your project description]"

For battle mode:

"Use skillpilot_plan in battle mode for [your project description]"

Note: GITHUB_TOKEN is optional but recommended — without it GitHub API limits you to 60 requests/hour. Create a token here (no scopes needed).

MCP Tools

| Tool | Description | |------|-------------| | skillpilot_plan | Main tool. One request → complete plan with agents, skills, projects, API keys, roadmap. Supports mode: "battle" for adversarial evaluation. | | skillpilot_team | Create full agent team with weapons and community skills | | skillpilot_team_estimate | Preview team size, roles, and skills before committing | | skillpilot_catalog | Browse 60+ curated MCP agents across 24 categories | | skillpilot_install | Generate .mcp.json, .env.example, .cursorrules, CLAUDE.md, ROADMAP.md | | skillpilot_explain | Detailed info about any agent or category | | skillpilot_discover | Search 37K+ MCP servers on GitHub | | skillpilot_skills | Find AI skills: universal prompts, IDE-specific, domain expert | | skillpilot_projects | Find open source projects to fork or reference |

14 Agent Roles

| Role | Weapons | Skills | |------|---------|--------| | Backend Architect | Supabase, Neon, System Architect, API Designer | Claude API, Node.js/Express, Hono, API Design | | Auth & Security Lead | Clerk, Semgrep, Security Expert | Supabase Auth, Better Auth, Security Audit, 007 | | Frontend Architect | Figma, Frontend Architect, UI/UX Expert | Next.js/React, shadcn/ui, WCAG Audit | | Payments Engineer | Stripe, Security Expert | Stripe Agent, Fintech Compliance, Billing | | AI/ML Engineer | Context7, System Architect | MCP Builder, Prompt Engineering, AI Toolkit | | Communications Engineer | Resend, Slack | Email Automation, Klaviyo, Marketing | | Business Analyst | Firecrawl, Business Analyst, System Architect | Market Sizing, Free Cash Flow, SWOT, Forecasting | | Strategy Consultant | Firecrawl, PostHog, Strategy Consultant | Porter's Five Forces, PEST, Customer Segmentation | | QA Lead | Playwright, QA Engineer | Webapp Testing, E2E Testing | | Code Review Lead | GitHub, Code Reviewer | Code Review Expert, Clean Code | | Performance Engineer | Upstash, Sentry, Performance Engineer | DevOps Troubleshooter, Cost Optimizer | | DevOps Lead | Vercel, Cloudflare, DevOps Engineer | Docker, Deployment Engineer, Pipeline Architect | | Analytics Engineer | PostHog, Data Engineer | Data Analyst, A/B Tests, dbt | | Content & CMS | Sanity, Firecrawl | Doc Co-Authoring, Content Strategy, Growth Engine |

8 Skill Sources (2,300+ community skills)

| Tier | Source | Skills | |------|--------|--------| | A | Anthropic Official | 17 | | A | GitHub Copilot | 362 | | A | Awesome CursorRules | 178 | | A | Peter Steinberger Rules | 15 | | B | Claude Skills Multi-Platform | 278 | | B | Workflow Skills | 66 | | B | Framework Prompts | 97 | | B | Business Strategy Planning | 50 | | C | Antigravity Skills | 1,340 |

21 Built-in Skills

Ready-to-use instructions across all categories:

• Auth — providers, sessions, protected routes, OAuth, roles
• Database — schema design, naming, indexes, migrations, RLS
• Payments — checkout flow, subscriptions, webhooks, error handling
• AI — streaming chat, context management, token optimization, RAG
• Design — component architecture, responsive, dark mode, accessibility
• Marketing — SEO, meta tags, landing page structure, schema.org
• Deploy — CI/CD pipeline, env vars, preview deployments, checklist
• Security — OWASP top 10, input validation, SQL injection, secrets
• Business Plan — TAM/SAM/SOM, unit economics, P&L projections, NPV/IRR, break-even
• Market Research — competitive intelligence, customer segmentation, industry analysis
• And more: email, analytics, monitoring, testing, cache, search, CMS, video, storage, notifications

Smart Features

• Agent Team System — parallel swarm of specialized agents with sub-agents and community skills
• Battle Mode — 3 competing strategies evaluated by your AI for higher quality plans
• 8 skill sources — 2,300+ community skills from Anthropic, GitHub Copilot, CursorRules, and more
• Business planning — financial modeling and market research agents with real web scraping
• LLM-driven categories — Your IDE's AI picks the right categories, no keyword guessing
• Stack enrichment — Redis in your stack? Cache skill auto-included. Python + FastAPI? AI skill added.
• GitHub discovery — Finds relevant MCP servers, skills, and projects beyond the built-in catalog
• Scoring algorithm — Ranks GitHub results by popularity, freshness, official status, license, activity

Development

npm install
npm run build
npm test          # 160 tests
npm run dev       # development mode

Tech Stack

• TypeScript + Node.js
• @modelcontextprotocol/sdk for MCP protocol
• GitHub API for discovery (37K+ MCP servers)
• Zod for input validation
• Vitest for testing

License

AGPL-3.0