Tickscope

Tickscope MCP

Real-time, free crypto market data for any AI agent — via MCP.

English · 한국어 · 中文 · 日本語

Tickscope is a self-hostable Model Context Protocol server that gives any MCP client (Claude Code, Cursor, Codex, Gemini CLI, …) real-time and historical crypto market data for free. It keeps exchange WebSocket connections warm in the background, so your agent reads prices that are fresh to the sub-second, straight from a live connection. The same server covers 73 technical indicators and chart-structure recognition, with no API keys.

⚠️ Educational/research tool. It does not provide financial, investment, or trading advice, and it does not guarantee data accuracy or timeliness.

---

Contents

• Why · See it run · 30-second install
• Supported exchanges · Tools · Indicators · Structure recognition
• Example prompts · Configuration · Development · Roadmap

Why

Trading agents are exploding, yet their data layer stays fragmented, REST-poll-only, and often locked behind a paywall. Tickscope gives those agents real-time, free market data from one server — many exchanges, no API keys.

See it run

uv run examples/demo.py            # live BTC/USDT walkthrough (no API keys)

uv run examples/demo.py ETH/USDT 4h

A colorized terminal walkthrough — cold→warm freshness (REST → WebSocket), indicators with signals, divergence, market structure, and support/resistance — against live Binance/Bybit/OKX. See examples/RECORDING.md to turn it into the GIF above.

30-second install

uvx tickscope-mcp

Register it with your client (Claude Code example, examples/claude_code_config.json):

{

"mcpServers": { "tickscope": { "command": "uvx", "args": ["tickscope-mcp"], "env": { "TICKSCOPE_EXCHANGES": "binance,bybit,okx", "TICKSCOPE_DEFAULT_EXCHANGE": "binance" } } } }

Cursor, Codex and Gemini CLI use the same command/args/env shape in their respective MCP config files.

Supported exchanges

Any ccxt-supported exchange can be enabled via TICKSCOPE_EXCHANGES. Public data only — no keys required.

Tools

Every market-data response includes source (websocket|rest), age_ms, and timestamp so the freshness is always provable.

Indicators (73)

• MAs / overlays: sma ema wma smma dema tema hma vwma zlema alma kama trima lsma vidya t3 vwap vwapbands bbands donchian keltner supertrend ichimoku psar
• Momentum: rsi stochrsi macd ppo stoch cci willr roc mom tsi ao cmo uo dpo trix coppock kst fisher rvi mfi wavetrend squeeze qqe crsi stc elderray zscore linregslope
• Volatility: atr natr stdev hv chop ulcer massindex
• Volume: obv adl cmf chaikinosc eom fi pvt vo klinger
• Trend: adx dmi aroon vortex
• Structure: heikinashi pivots

Specs are "name:p1,p2" and also accept Pine Script syntax — ta.rsi(14), ta.ema(20), ta.wt(10,21), ta.sqz — so TradingView users can paste familiar expressions. Derived signals include overbought-oversold state, MACD/PPO/WaveTrend/QQE cross, zero-line cross for oscillators, Supertrend/PSAR direction & flip, squeeze on/off, DMI/Heikin-Ashi trend, and Ichimoku cloud position. Includes crypto/Pine favorites (WaveTrend, TTM Squeeze, QQE, Connors RSI, Schaff Trend Cycle, VIDYA, T3). Adding a new indicator is a one-line REGISTRY declaration.

Structure recognition

On top of numeric indicators, Tickscope describes *what the chart is doing*: detect_patterns names candlestick patterns (engulfing, hammer/hanging man, doji family, morning/evening star, three soldiers/crows, …) with their bias; analyze_structure returns swing highs/lows labeled HH/HL/LH/LL, the inferred trend, and Break-of-Structure / Change-of-Character events (SMC-style); find_support_resistance clusters swing pivots into support/resistance zones with touch counts. These give an agent the vocabulary to describe a chart the way a trader would.

Deep analysis

deep_analyze answers a question about a symbol in one call, instead of making the agent chain a dozen tools. It returns:

• Multi-timeframe trend confluence — the same symbol read across a 1d/4h/1h ladder, with whether the timeframes agree or conflict.
• Market-state context — where price sits in its recent range (percentile), the trend state (trending_up / trending_down / ranging, from ADX + Kaufman efficiency ratio), and the volatility state (from ATR percentile), so a bare "RSI 30" reads against the conditions it showed up in.
• Historical signal performance — for the current divergence, the forward-return distribution of every past *confirmed* occurrence on this symbol/timeframe (count, win rate, median). A strictly causal event study — no look-ahead, no repaint.
• A synthesized verdict — bias, confidence, timeframe agreement, the execution-timeframe market state, and explicit caveats, all computed deterministically in Python so the call never hinges on the model eyeballing raw numbers.

compute_indicators now carries the same market-state context inline (it's ~free), and signal history is memoized per closed bar, so warm reads stay fast. Clients that support MCP prompts expose this as a slash command — /mcp__tickscope__deep_analyze (symbol + timeframe) — to trigger a full read on demand.

Resources

Supporting clients can also read live state as MCP resources: tickscope://status, tickscope://watched, and the template tickscope://ticker/{exchange}/{symbol}.

Example prompts

• "What's BTC/USDT trading at on Binance right now, and the 24h change?"
• "Should I care about this SOL/USDT setup? Do a deep, multi-timeframe read."
• "Is there an RSI divergence on BTC/USDT 4h — and how has that signal performed on this pair before?"
• "Screen the top 50 USDT pairs by volume for ones with RSI below 30."
• "Map the key support/resistance on ETH/USDT 4h and tell me if the structure just broke (BOS/CHoCH)."
• "What's the perpetual funding rate for BTC on Bybit right now — is it overheated?"
• "Compare BTC/USDT across Binance, Bybit and OKX and show the arbitrage spread."

See examples/demo_prompts.md for a fuller recipe book (deep analysis, signal backtests, screening, monitoring, strategy checks).

Configuration

All settings are environment variables (see .env.example):

Development

uv venv && uv pip install -e ".[dev]"

pytest # ~100 unit + MCP-integration tests (live excluded) pytest -m live # live exchange tests (Binance/Bybit/OKX, run locally) ruff check . && mypy # lint + type gates

Tests cover indicator math against reference values, service cache/auto-watch logic, the full MCP tool path (tests/test_mcp_integration.py calls tools through mcp.call_tool), price-structure recognition, and a live suite that exercises the whole stack against real exchanges. See CONTRIBUTING.md for the project layout and contribution flow.

Roadmap

• Pine Script-style indicator mapping (ta.rsi, ta.crossover, …)
• 73 indicators + candlestick patterns + market structure (BOS/CHoCH)
• Multi-exchange aggregation (weighted price / spread)
• MCP resource push for watched symbols
• Anchored / session VWAP
• More exchanges (Kraken, Bitget, Gate, …)
• Agent Skill (SKILL.md) wrapper

Contributing

Issues and PRs welcome — see CONTRIBUTING.md and our Code of Conduct. Keep dependencies minimal and the v1 scope read-only (public data, no order execution, no API secrets).

License

MIT © Tickscope contributors.

Disclaimer

This tool is for educational and research purposes only. It is not financial, investment, or trading advice. Market data may be delayed, incomplete, or inaccurate; do not rely on it for real trading decisions. Respect each exchange's terms of service and rate limits. See SECURITY.md.