Strike research tools

This skill is optional. The desk executes through the signed Strike REST API (strike-orders) and reads markets from the public Price Service (strike-market-data). It needs nothing here to trade.

What the crowdtime MCP adds is a handful of things Strike's own API does not compute: a liquidity screen, ready-made technical indicators, research playbooks, and a way to reach the user outside the chat. If it is not connected, say so when a request would have used it and fall back to primary sources; never let its absence become a silent gap.

Nothing here executes. The desk deliberately does not use the MCP's order tools: two write paths is exactly what the one-writer rule exists to prevent. Execution is the signed REST API, always.

1. Connecting (optional)

Two routes. Grok connector: grok.com/connectors -> New Connector -> Custom -> the URL above. Provisioned at team level by an admin, and xAI does not document whether a connector reaches a named Bot, so test it. Desk computer: JSON-RPC over HTTPS, which always works.

call_mcp() {                     # usage: call_mcp <tool> <json-arguments>
  curl -sS --max-time 60 https://mcp.crowdtime.io/mcp \
    -H "Authorization: Bearer $STRIKE_MCP_TOKEN" \
    -H 'Content-Type: application/json' \
    -H 'Accept: application/json, text/event-stream' \
    -d "$(printf '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"%s","arguments":%s}}' "$1" "$2")" \
  | python3 -c 'import sys,json
d=json.load(sys.stdin)
if "error" in d: print("MCP ERROR:", json.dumps(d["error"])); raise SystemExit(1)
for c in d["result"]["content"]:
    if c.get("type")=="text": print(c["text"])
print("isError:", d["result"].get("isError", False))'
}

The token goes in Grok Bot's secure secret store as STRIKE_MCP_TOKEN, never in chat or a file. It is a second credential with trading power, so if the desk is not using the research tools, do not provision it at all - the smallest credential set is the safest one.

Symbols differ here. The MCP speaks -PERP (ADA-PERP, and GOLD-PERP for what the Price Service calls XAU-USD); everything else on this desk speaks -USD. It also covers only fifteen of Strike's thirty-one markets, so a market can be tradeable and still have no MCP coverage.

2. Liquidity screen

call_mcp strike_scan_markets '{"days":7}'

Ranks the markets it covers by a composite "hotness" score and returns an excluded list naming those below its open-interest and volume floors.

Hotness is a screen, not a signal. It has no direction, and a market tops it as readily for being violently liquidated as for being accumulated. The excluded list is the genuinely useful half: it says which markets are too thin to trade before anyone reads a chart. Leave min_oi_notional_usd and min_avg_daily_volume_usd unset unless the user names a figure.

3. Computed indicators

call_mcp strike_get_market_snapshot '{"symbol":"ADA-PERP","interval":"1h"}'

Mark, last close, RSI(14), EMA(20/50/200), MACD, Bollinger(20,2), ADX(14), ATR(14), and a trend label. Returns isError when its upstream feed is stale - then the indicators are unavailable and the desk falls back to /v2/klines and says it did.

An indicator is a fact about recent closes. RSI(14) at 78 is a fact; "so it will fall" is not this desk's job, and no Bot on the floor says it.

4. Research playbooks

call_mcp crowdtrendz_crypto_news_research '{"symbol":"ADA","horizon":"next 7 days","lookback":"last 14 days"}'
call_mcp crowdtrendz_stock_news_research '{"symbol":"NVDA","horizon":"next earnings"}'
call_mcp crowdtrendz_stock_dividend_research '{"exchange":"NASDAQ","symbol":"MU","horizon":"next 30 days"}'

These return instructions, not findings. Calling one produces no research: it hands back a playbook, and the Bot must then do the web research itself and write the report under the rules it was given. Never present a returned playbook as findings, and never say "research shows" on the strength of having called the tool. The citations in the report are pages you read, with links and UTC times.

symbol is a plain ticker (ADA, NVDA), not a Strike perp symbol. Ask the user for horizon and lookback in one short question when they have not said - a report whose sources predate the last move is worse than none. strike_review_trading_style behaves the same way for the account's own history, and is advisory: its suggestions never become trades except through the lifecycle.

Strike lists equities and commodities, so scheduled earnings, guidance, index events and ex-dates matter here. Put dated events in /workspace/trading-desk/research/calendar.md with a source link and a UTC time. A perp on an equity is not the equity: it pays no dividend and can trade when the underlying market is shut. Say which you are describing.

5. Bodega prediction markets

call_mcp bodega_list_markets '{"status":"open","sort":"volume","limit":25}'

A Cardano prediction-market protocol; each market is a yes/no question with Yes/No prices as 0-1 implied probabilities. Read-only, and the desk does not trade it - there is no execution path and none is improvised.

Its use is a second, market-priced read on a question the desk cares about. Quote the implied probability with the market's depth beside it: 0.78 on a market with 400 ADA of volume is an opinion, not a price. open and closed are decided by the deadline, not the upstream status field. Treat a Bodega price as reported evidence, never verified fact.

6. Discord alerts

call_mcp discord_send_message '{"content":"SG-20260910-01 filled: long 4800 ADA-USD @ 0.2101. Stop resting at 0.1995.","connection":"Trade Alerts"}'

This leaves the conversation, so: only when the user has asked for alerts, to a webhook they configured; never a credential, an account identifier or a balance they have not agreed to have leave the chat; under 2000 characters.

An alert is a notification, not an instruction and not an approval. Nothing posted to Discord can authorise a trade, and no reply there reaches the desk.

7. Evidence tiers

Label every claim: verified (you read it at the primary source and linked it), reported (a credible outlet says so; a Bodega market prices it), claimed (someone on social media says so), inferred (your reasoning). Never promote a claim up a tier without new evidence. Missing information is unknown, not "probably fine".