MCP for AI trading agents

How do AI agents get real-time crypto trade calls?

Q: How does an agent request a call?

The agent calls one MCP tool, get_trade_call, with a coin, timeframe, and venue. It connects over remote HTTPS, a local stdio process, or an MCP registry. No chart, no manual interpretation.

Q: What's in the response?

A composite verdict: direction (BUY, SELL, or HOLD), a confidence score, the current market regime, and the underlying factors (funding, open interest, trend persistence, breakout state). The agent decides whether to act.

Q: What cadence are calls fired at?

Calls are evaluated on a five-minute-or-slower decision cadence; the agent can request a fresh verdict on demand at any supported timeframe. Sub-five-minute execution is a downstream concern handled by the agent.

Q: Can the agent verify the record itself?

Yes. Every call is hashed and published to a Merkle root on Base L2 before its outcome is known. The agent can recompute a call's hash and check it against the on-chain root at algovault.com/verify. The live win rate is at algovault.com/track-record.

AI agents get crypto trade calls by calling an MCP server. AlgoVault returns one composite verdict per call: direction, confidence, and market regime, across five perpetual-futures venues. It posts a 91.6% PFE win rate over 233,000 verified calls, each Merkle-anchored on Base L2. The agent reads the verdict and decides execution.

How does an agent request a call?

The agent calls one tool over the Model Context Protocol. No chart, no adapter glue. Connect over remote HTTPS, a local stdio process, or an MCP registry, then call get_trade_call with a coin, timeframe, and venue.

POST https://api.algovault.com/mcp        # tool: get_trade_call
{ "coin": "ETH", "timeframe": "15m", "exchange": "HL" }

-> { "call": "SELL", "confidence": 64, "regime": "TRENDING_DOWN",
     "factors": { "funding": "ELEVATED", "trend_persistence": "MEDIUM",
                  "breakout_pending": "INACTIVE" } }

What's in the response?

One verdict the agent can act on, not raw series to interpret:

call — BUY, SELL, or HOLD.
confidence — a score the agent can threshold on.
regime — TRENDING_UP, TRENDING_DOWN, RANGING, or VOLATILE.
factors — funding state, open-interest momentum, trend persistence, breakout state.

HOLD calls are always free, so an agent can poll continuously across 5 venues without burning quota on no-trade reads.

How fresh are the calls?

An agent requests a fresh verdict on demand at any supported timeframe. The track record is built on a five-minute-or-slower decision cadence; faster execution is the agent's downstream job. Coverage spans a broad asset set; the live counts are on the track record.

How does the agent verify the record?

Every call is hashed and published on Base L2 before its outcome is known. The agent can recompute any call's hash and check it against the on-chain Merkle root at /verify. The record is audited, not trusted.

FAQ

How does an agent request a call?

One MCP call to get_trade_call with a coin, timeframe, and venue, over remote HTTPS, stdio, or a registry.

What's in the response?

Direction, confidence, market regime, and the underlying factors. The agent decides whether to act.

What cadence are calls fired at?

A five-minute-or-slower decision cadence; the agent can pull a fresh verdict on demand.

Can the agent verify the record itself?

Yes — recompute a call's hash and check it on-chain at /verify; the live win rate is at /track-record.

Verify it yourself

See the live track record, verify any call on-chain at /verify, or read how it works. Built by AlgoVault Labs.

Run get_trade_call free — 100 calls/month → See the live track record →

This is call interpretation, not investment advice; agents decide execution.