AI Starter Protocol — MCP Server

MCP server that helps AI coding assistants find the best starter repo for any task. When you ask Claude "build me a web scraper" or "set up Stripe webhooks", this server returns the best matching starter repo with quickstart instructions.

Install

python3 -m pip install mcp

Add to Claude Code

Add to ~/.claude/settings.json:

{
  "mcpServers": {
    "ai-starters": {
      "command": "python3",
      "args": ["/path/to/ai-starter-protocol/mcp-server/server.py"]
    }
  }
}

Tools

find_starter(query: string)

Natural language search. Returns top 3 matching starters ranked by relevance with confidence scores.

find_starter("scrape a website")
→ Returns python-web-scraper with quickstart command

find_starter("stripe webhooks")
→ Returns stripe-webhook-handler

list_starters(language?: string)

List all available starters. Optionally filter by language.

list_starters()           → all starters
list_starters("python")   → python starters only

get_starter_details(repo: string)

Full details for a specific starter including quickstart, dependencies, and related projects.

get_starter_details("LuciferForge/python-web-scraper")

Registry

The server reads from ../registry.json by default. Override with:

export STARTER_REGISTRY_PATH=/path/to/registry.json

Community Starters

The registry supports community contributions. Add a community_registry_url field to your registry.json pointing to another registry, and those starters will be merged in automatically.

{
  "community_registry_url": "https://example.com/community-starters.json",
  "starters": [...]
}

Or set via environment variable:

export COMMUNITY_REGISTRY_URL=https://example.com/community-starters.json

How Matching Works

1. Query is tokenized into keywords (stop words removed)
2. Each starter scored: +3 for keyword in recommend_when, +2 in purpose, +1 in repo name
3. Top 3 returned if score >= 2
4. Confidence: high (10+), medium (5+), low (2+)