@bitget-ai/bitget-signal

The market-signal product of the Bitget Agent Hub — a skill + MCP bundle that gives Claude Code, Codex, and OpenClaw deep crypto market intelligence: macro, on-chain, sentiment, technicals, and news. No API key. No account. No setup beyond npx.

npx @bitget-ai/bitget-signal

Prerequisites: Node.js ≥ 20. Python with pandas and numpy only if you use the technical-analysis skill (pip install pandas numpy). Everything else is bundled markdown plus an auto-configured public MCP server.

What you get

Five purpose-built market-analysis skills. Each carries explicit trigger phrases so your AI assistant invokes the right one without you needing to be technical.

| Skill | Triggers when you ask… | Powered by | |---|---|---| | macro-analyst | "is the macro good for crypto?", "Fed rate impact on BTC", "BTC vs DXY", "yield curve" | Fed policy, FOMC news, yield curve, cross-asset correlation (BTC vs Gold/DXY/Nasdaq/S&P/10Y) | | market-intel | "are whales buying?", "ETF flows", "where are we in the cycle?", "is it accumulation zone?" | On-chain capital flows, ETF data, DeFi TVL, cycle indicators (AHR999, Pi Cycle, rainbow) | | sentiment-analyst | "is the market greedy or fearful?", "are longs crowded?", "funding rate", "squeeze risk?" | Fear & Greed Index, long/short ratio, OI, funding, taker ratio, Reddit social signals | | technical-analysis | "is BTC overbought?", "RSI on ETH", "support/resistance", "4h setup" | 23 indicators across 6 categories (Trend, Volatility, Oscillator, Volume, Momentum, S/R) | | news-briefing | "what's happening in crypto?", "morning briefing", "any news on X?", "what's the catalyst?" | 44 RSS/Atom feeds, social trending boards, narrative synthesis |

Roadmap — Bitget-only signals

Beyond the generic skills above, bitget-signal is the launch vehicle for signals only Bitget can produce — built directly from Bitget's exchange data:

• top-trader-flow — aggregated, anonymised positioning of Bitget's top copy-trading leaders.
• derivatives-structure — perp basis, term structure, and funding-rate curve from Bitget's orderbook.
• large-flow-detect — whale-sized order detection on Bitget pairs in real time.

These will ship as additional skills inside this same package — no new install required. Watch Bitget-AI/agent-hub for release notes.

Install

# Default — Claude Code skills + register the bitget-signal MCP server
npx @bitget-ai/bitget-signal

# All supported AI tools (Claude Code + Codex + OpenClaw)
npx @bitget-ai/bitget-signal --target all

# Specific targets
npx @bitget-ai/bitget-signal --target claude,codex

# Interactive picker
npx @bitget-ai/bitget-signal --interactive

What npx @bitget-ai/bitget-signal does, in order:

1. Copies the five skill folders into your AI tool's skill directory.
2. Registers the public bitget-signal MCP server (HTTP transport) in your tool's MCP config.

After installing, restart your AI host (e.g. Claude Code) so the new skills and MCP server are picked up.

| Target | Skill location | MCP config | |---|---|---| | Claude Code | ~/.claude/skills/ | via claude mcp add | | Codex | ~/.codex/skills/ | written to ~/.codex/config.toml | | OpenClaw | ~/.openclaw/skills/ | written to ~/.openclaw/config.json |

How it differs from @bitget-ai/bitget-agent-skill

| | bitget-agent-skill | bitget-signal | |---|---|---| | Purpose | Trade on your Bitget account | Read the market | | API key required? | Yes (your Bitget credentials) | No | | Stack | Trading Stack | Signal Stack | | Triggers on | "buy / sell / show my positions / cancel order" | "is BTC overbought / Fed rate / whales / sentiment" | | Backend | Bitget REST API (private endpoints) | Public market-data MCP server |

The two are complementary and can be installed side by side. Many users want both: an AI that reads the market and executes accordingly.

What's a skill, what's the MCP server?

bitget-signal is two things bundled:

• The five skill files — markdown that teaches your AI assistant what each skill does and when to invoke it. These are pure text, no runtime.
• An MCP server registration — your AI host gets pointed at Bitget's public market-data MCP (HTTP transport). When the AI calls a tool, the request goes to the public server; results come back. No proxy through your machine, no credentials needed.

Together: the skills are the prompt, the MCP server is the tools. Either alone is half the product; the installer wires both up in one command.

Migration from old package names

- npm install -g bitget-skill-hub          # original name (deprecated)
- npm install -g bitget-ai-analyst         # interim name (deprecated)
+ npx @bitget-ai/bitget-signal             # canonical name

If you previously installed bitget-skill-hub or bitget-ai-analyst, also remove the old market-data MCP entry from your AI host config — the new install registers the server under the name bitget-signal.

Manage with the meta installer

If you're using more than one Bitget AI package, the meta installer can manage everything at once:

npx @bitget-ai/bitget-agent-installer

License

MIT

Part of the Bitget Agent Hub ecosystem · Signal Stack. Trading Stack: agent-sdk · agent-cli · agent-mcp · agent-skill