Let an AI agent run
your market watch.
Airohead speaks the Model Context Protocol, so Claude — or any MCP-capable agent — can read live market data, build and arm hunts, replay a draft hunt’s triggers over recorded history and read the measured results back. The agent acts as your account and gets exactly your permissions.
Reads the market
Live prices, funding, volatility and edge across every tracked instrument. Free, and the same rows the dashboard renders.
Builds and arms hunts
Multi-condition hunts with an optional trade leg — direction, take-profit, stop-loss — evaluated on paper by the same engine.
Reports honestly
Track records come from real closed positions, and a simulated response is flagged as simulated so an agent cannot overclaim.
Connect it in three steps.
Airohead is hosted, so there is nothing to install and no runtime to keep alive.
Create an API key
Open Settings → API keys and create one. The key is shown once — Airohead stores only a SHA-256 hash, so it cannot be recovered or read back out of the database. Lose it and you make a new one.
A key inherits the permissions of the account that made it, and nothing more. It also cannot mint further keys, so a leaked agent credential cannot quietly widen its own access. Revoke any key from the same screen and it stops working immediately.
Point your agent at the endpoint
In Claude Code, one line connects it:
claude mcp add airohead --transport http https://your-airohead-host/api/mcp \ --header "Authorization: Bearer airo_sk_your_key_here"
For Claude Desktop, Cursor, or anything else that takes a custom remote MCP connector, add the same endpoint:
{
"mcpServers": {
"airohead": {
"type": "http",
"url": "https://your-airohead-host/api/mcp",
"headers": {
"Authorization": "Bearer airo_sk_your_key_here"
}
}
}
}Confirm it worked by asking your agent to call whoami — it should name your account.
Prefer to run it locally? (stdio)
If you would rather the agent spawn the server as a local process, clone the repo and point at the stdio entry instead. Both transports serve the identical tools from one registry.
{
"mcpServers": {
"airohead": {
"command": "bun",
"args": ["run", "/absolute/path/to/mcp/airohead-mcp.ts"],
"env": {
"AIROHEAD_API_KEY": "airo_sk_your_key_here",
"AIROHEAD_URL": "https://your-airohead-host"
}
}
}
}That path is dependency-free — it speaks JSON-RPC over stdio directly, so bun is all you need. Change AIROHEAD_URL to point at a local dev server instead.
Ask it something
Once connected, talk to your agent normally. Things that work today:
- “What is moving in Airohead right now, and which has the highest edge score?”
- “Arm a hunt for me: NVDA/USDG edge above 80 and 24h change positive.”
- “Build a long leg on WETH/USDG with a 3% take-profit and a 1.5% stop.”
- “Replay that idea's triggers over the last 30 days before arming it.”
- “How are my hunts with a trade leg actually doing?”
Every tool is free today.
There are no tiers and no agent surcharge, and nothing on this surface is billed yet. When metering arrives it lands on the calls that run a model, and nowhere else.
Every tool listed below, ask_airo included. Reads, hunts, alerts, portfolio, trigger replay and whoami cost nothing and are not counted against a balance, because there is no balance to count against yet.
ask_airo is the one call that runs a model, so it carries a burst limit of six a minute per account. That limit is the only thing standing between an agent and the analyst — no charge, no quota, no cap.
A key cannot sign a transfer mid-request, so the metered design draws on a prepaid USDG balance inside a per-key cap. None of that is built — it is the plan, described here so the shape is not a surprise later.
The per-query price is not set, so this page does not print one — see pricing for the mechanism as designed. Until it ships, an agent calling Airohead is not spending anything, and this page will say so rather than imply a bill that does not exist.
Running alongside your other tools.
MCP is an open standard and an agent can hold several connections at once. If your agent is already connected to a brokerage’s MCP server, Airohead sits beside it rather than competing with it: Airohead is where the agent does the finding and the reasoning, and your broker stays the only thing that can touch a real order.
That boundary is enforced here, not just described. Airohead has no order-placement tool and no custody of funds — create_hunt opens paper positions the engine tracks for a measured record, and nothing in this API can move money.
Airohead is not affiliated with, endorsed by, or connected to any broker. Anything an agent does with a real broker is between you and that broker — decide for yourself what you let an agent execute.
What the agent can call.
Every tool below is registered on the server. No count is printed here on purpose — the catalog moves, and tools/list is always the authoritative answer.
list_marketsFreeEvery instrument Airohead tracks — price, 24h change, high/low, volume, volatility, perp funding and the 0–100 edge score.
Start here. Other tools need these exact symbols.
market_regimeFreeThe regime engine's read on every tracked symbol — trending, mean-reverting, high-volatility or illiquid — with confidence and the metrics behind it.
Deterministic arithmetic over recorded history. A symbol the engine declines to label comes back as null with the reason.
market_precisionFreeThe precision board: how well each tracked market can be measured right now, ranked, with the inputs that were unavailable.
A missing input scores zero and keeps its weight — a half-measured 50 is never dressed as a 100.
ask_airoFree · metering plannedAsk AIRO, the built-in analyst, a question in plain language. It answers only from Airohead's tracked instruments and will not invent a metric it cannot read.
The one call that runs a model. It is free today and throttled to six a minute; when the metered rail lands, this is the tool it lands on.
list_huntsFreeThe account's hunts grouped by armed / triggered / history, each with the plain-language narration of what it evaluates.
create_huntFreeCreate a hunt. Alert-only, or with a trade leg carrying direction, take-profit and stop-loss.
Paper only — no real funds move and no exchange order is placed.
pause_huntFreePause an armed hunt so the engine stops evaluating it.
delete_huntFreeDelete a hunt permanently.
Irreversible — a well-behaved agent should confirm with you first.
backtest_huntFreeReplay a draft hunt's entry conditions against collected history to count how often they would have fired, before arming it.
Entry-only. It counts triggers over the history Airohead actually recorded — it does not simulate the trade leg, exits or a P&L curve.
hunt_performanceFreeMeasured track record for hunts with a trade leg: win rate, realized P&L, and open positions marked to the live price.
Computed from real closed positions — never self-reported.
list_alertsFreeAlerts received, each with the briefing: thesis, evidence, risk and the invalidation level.
get_portfolioFreeEquity, realized and unrealized P&L, Sharpe, hit rate, drawdown and asset exposure.
Zeros are honest zeros, not errors.
whoamiFreeWhich Airohead account the key is acting as. Use it to confirm setup.
Or skip MCP entirely.
The same API key authenticates any HTTP request, so a cron job, a bot or your own code can call the endpoints directly.
Send the key as Authorization: Bearer or X-Airohead-Key.
curl https://your-airohead-host/api/trpc/market.getAll \ -H "Authorization: Bearer airo_sk_your_key_here"
Airohead stores numbers as integers to avoid floating-point drift. Agents get this wrong unless told, and the tool schemas repeat it — but it is worth knowing:
What an agent cannot do.
Move real money. A hunt's trade leg opens paper positions only — nothing is sent to an exchange and your wallet is never touched.
Mint more API keys. Key management requires a real signed-in session.
See anyone else's data. A key resolves to one account and carries exactly its permissions.
Hammer the analyst. Model calls are capped at six a minute per account, and that throttle applies to a key exactly as it applies to you.
Outlive your trust. Revoke a key and the next request fails immediately.