rs-radar

Relative-strength & momentum radar for crypto and US stocks, as an MCP server.

Ask your AI assistant "what's strong vs BTC right now?", "who led the bounce since 20:24?" or "start the bot and ping me on high-confidence setups". rs-radar does the market-data plumbing and the statistics; your MCP client (Claude Code, Claude Desktop, Cursor, …) does the conversation.

• Relative strength that means something. Every "vs benchmark" figure is beta-adjusted: a coin that normally moves 2× BTC gets no credit for rising 2% when BTC rises 1%.

• Signals, not noise. A setup must clear hard gates (outperformance, a ≥3σ move for that instrument, a volume surge, an intact uptrend) before it's scored, and alerts fire only at high confidence.

• Trade plans included. Entry, max entry, structure/ATR stop-loss, and 1.5R / 3R targets.

• A bot that outlives your chat. Desktop notifications (macOS/Linux) and optional Discord/Slack webhooks, with cooldowns and a health heartbeat.

• Watchlists from screenshots. Drop an image of your exchange watchlist into the chat; your assistant reads the tickers and rs-radar validates each against live data.

• No API keys required. Crypto: CoinGecko + Binance public data (Crypto.com fallback). US stocks: Yahoo Finance chart data, benchmarked against SPY.

$ rs-radar since --top 5
crypto | BTC 84640.5 | 5m -0.00%  15m +0.06%  1h +0.04%  4h +0.06% | as of 2026-10-03T08:50+00:00
anchor 2026-10-03T07:59+00:00 | benchmark +0.10% since | 47/88 beating
symbol  change_pct  vs_benchmark_pct  beta_adjusted_pct  beta  volume_ratio
    AR        7.61              7.52               7.45  1.73          77.8
   ZRO        4.43              4.34               4.14     3           2.5
  KITE         2.3              2.21               2.24  0.68             1
   WLD        1.99              1.89                1.7     3           1.5
  MINA        1.46              1.36                1.3  1.65           1.4

Quick start

rs-radar runs locally via uv. Nothing to clone; uvx fetches and runs it.

Claude Code

claude mcp add rs-radar -- uvx --from git+https://github.com/ajinkya-cs/rs-radar rs-radar serve

Claude Desktop, Cursor and other MCP clients: add to the client's MCP config (claude_desktop_config.json, .cursor/mcp.json, …):

{
  "mcpServers": {
    "rs-radar": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/ajinkya-cs/rs-radar", "rs-radar", "serve"]
    }
  }
}

Then just ask: "Scan crypto for relative strength", "RSI on SOL, ETH and AR on the 4h", "Give me a trade plan for NVDA", "Start the bot for crypto and stocks".

Related MCP server: Rozkoduj MCP

Slash commands

The server ships MCP prompts, which clients expose as slash commands. In Claude Code they appear as /mcp__rs-radar__<name>:

Tools

Every tool takes market: "crypto" | "stocks". Crypto covers the top 200 by market cap (stablecoins, wrapped and liquid-staking duplicates removed) against BTC; stocks cover ~100 US large caps against SPY.

The bot

rs-radar bot start --market crypto --market stocks --min-confidence 85
rs-radar bot status
rs-radar bot stop

Or from chat: "start the bot on my watchlist at confidence 90". The bot runs as a detached process, so it keeps running after you close the MCP client. Each cycle it:

1. checks the market is live with one cheap request, so a closed stock market costs nothing;

2. scans the universe and/or your watchlist;

3. alerts on the single best new setup per market, then puts that symbol on cooldown (default 2h, persisted across restarts);

4. writes a heartbeat, so bot status can tell running from healthy.

Notifications look like LONG AR · 88/100 / Entry 4.583 SL 4.3235 TP 4.9673 / 5.3536. Set RS_RADAR_WEBHOOK_URL to also post to Discord or Slack.

How signals work

Momentum is measured on 1-minute bars aligned to the benchmark's clock. For each instrument:

Hard gates must pass (crypto: ≥ 1.5pp excess, ≥ +1% raw, ≥ 3σ, ≥ 2× volume; stocks use 0.5pp / +0.4%), then confidence (0–100) rewards independent confirmations. Signals are suppressed whenever data isn't live or the window spans a session gap, so an overnight stock gap can't masquerade as a 5-minute breakout. Full details and rationale: docs/methodology.md.

CLI

Everything is also available in the terminal:

uvx --from git+https://github.com/ajinkya-cs/rs-radar rs-radar scan              # market overview
rs-radar signals --market stocks --min-confidence 80
rs-radar rsi btc eth sol --interval 1d
rs-radar since --at 20:24                                                         # strength since a time
rs-radar plan NVDA --market stocks
rs-radar watchlist add sol ar zro && rs-radar watchlist show

Add --json to any analysis command for machine-readable output.

Configuration

All optional, via environment variables (set them in your MCP client's env block):

Keys are read from the environment only and never written to disk or logs.

Data sources & limitations

• Crypto: CoinGecko for the market-cap universe; Binance public market data for candles, with Crypto.com Exchange as fallback. Coins listed on neither (e.g. some exchange tokens) are skipped and reported.

• US stocks: Yahoo Finance's public chart endpoint. It's unofficial, free and keyless, but has no SLA. It's fine for personal research; for production use, implement a provider for a licensed feed (see CONTRIBUTING.md). Yahoo has no 4h bars, so stocks use 1h for trend.

• Public endpoints rate-limit. All requests share one client with bounded concurrency and backoff.

• Tested on macOS and Linux. The core works on Windows; desktop notifications there are not yet supported.

Architecture

flowchart LR
    client["MCP client<br/>(Claude, Cursor, …)"] -- stdio --> server["server.py<br/>tools + prompts"]
    cli["cli.py"] --> scanner
    server --> scanner["scanner.py<br/>snapshots · signals · RSI"]
    server --> bot["bot.py<br/>detached process"]
    bot --> scanner
    scanner --> signals["signals.py<br/>pure scoring & plans"]
    scanner --> providers["providers/<br/>crypto · stocks"]
    providers --> http["http.py<br/>retries · rate limits"]
    bot --> notify["notify.py<br/>desktop · webhook"]

Analytics (indicators.py, signals.py) are pure functions with no I/O; providers implement one small Protocol, so adding a market is one file plus tests.

Development

git clone https://github.com/ajinkya-cs/rs-radar && cd rs-radar
uv sync
uv run pytest                      # unit + in-process MCP tests, no network
uv run ruff check . && uv run mypy # lint & strict type-check

See CONTRIBUTING.md for conventions and how to add a provider or signal.

Roadmap

• Alpaca provider (free real-time IEX data with a key) as an alternative to Yahoo

• Windows toast notifications and a Telegram notifier

• Short setups (relative weakness) alongside longs

• Backtest harness to calibrate confidence thresholds on recorded data

Support

rs-radar is free and MIT-licensed. If it helped you catch a move, a tip keeps it maintained. ☕

Please double-check the network before sending: funds sent on the wrong network can't be recovered. Stars, bug reports and PRs are just as appreciated.

Disclaimer

rs-radar is a research tool. Its output is statistical observation, not financial advice. Markets can and do move against any setup. Size positions so that being wrong is affordable.

License

MIT