🛡️ Rift Analyst — League of Legends AI Agent

A conversational AI agent that analyzes League of Legends players and answers current-patch meta questions. Built with the Vercel AI SDK and Claude Haiku 4.5, deployable to Vercel and runnable locally.

🔗 Live demo: league-chatbot.vercel.app

Ask it: "Analyze Faker#KR1 on kr", "Who counters Darius top?", "Best build and runes for Jinx this patch?"

What it does

The agent has two complementary tool sets and decides which to call:

Capability	Data source	Why
Player profile, rank, win rate, recent match stats	Official Riot Games API (account-v1, summoner-v4, league-v4, match-v5)	Ground truth for player data — everything else is built on top of it
Champion meta, builds, runes, items, counters, tier lists (current patch)	OP.GG MCP server (mcp-api.op.gg/mcp)	Free, official, AI-native; aggregates millions of ranked games

Why these sources

• Riot API over scraping op.gg/u.gg: op.gg and u.gg both sit on top of the Riot API. Going direct gives authoritative data, no ToS risk, and real API engineering (auth, region routing, rate limits).
• OP.GG MCP over scraping lolalytics: lolalytics has no stable public API; community scrapers are unmaintained and break on layout changes. OP.GG's MCP server exposes the same kind of meta/build/counter data through a sanctioned, structured interface designed for agents.

Architecture

Browser (useChat UI)
   └── POST /api/chat
         └── streamText(model: claude-haiku-4-5)  ← Anthropic API
               ├── Riot tools      src/lib/riot.ts   (custom, typed)
               └── OP.GG MCP tools src/lib/opgg.ts   (discovered at runtime)

The model runs a tool loop (up to 12 steps): e.g. for a player analysis it calls lookupSummoner → getMatchHistory, then synthesizes the result.

Run locally

npm install
cp .env.example .env.local   # then fill in the keys
npm run dev                  # http://localhost:3000

You need two keys:

1. RIOT_API_KEY — generate at https://developer.riotgames.com (dev keys expire every 24h; request a production key for permanent hosting).
2. ANTHROPIC_API_KEY — from https://console.anthropic.com. Powers the Claude Haiku 4.5 agent.

The OP.GG MCP server needs no key. If it's unreachable, the agent degrades gracefully and still answers player-stats questions via the Riot API.

Run with Docker

The app builds to a standalone Next.js server in a small Alpine image.

docker build -t rift-analyst .
docker run -p 3000:3000 --env-file .env.local rift-analyst

On macOS without Docker Desktop, use a headless daemon:

brew install docker colima && colima start

Deploy to Vercel

npm i -g vercel
vercel            # preview deployment
vercel --prod     # production

Add RIOT_API_KEY and ANTHROPIC_API_KEY in the Vercel dashboard (Project → Settings → Environment Variables), or via vercel env add <NAME> production.

Stack

Next.js 16 (App Router) · Vercel AI SDK v6 · Claude Haiku 4.5 (Anthropic API) · Model Context Protocol (OP.GG) · TypeScript · Tailwind CSS · Docker.