A crypto MCP server with a graded record

Endpoint

Tools (12)

The list below is generated from the list the server answers tools/list with, so it cannot drift from what an agent sees. A paid tool takes a USDC payment on Base: pay the price, then pass the transaction hash as x402TxHash in the tools/call params. Only a transfer confirmed on Base counts. A Premium session runs every tool.

Add it to your client

Claude Desktop

claude_desktop_config.json. Claude Desktop starts local commands, so the mcp-remote bridge connects it to the remote URL (needs Node.js):

{
  "mcpServers": {
    "assistanthub": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://rmassistanthub.io/mcp"]
    }
  }
}

Cursor

~/.cursor/mcp.json, or .cursor/mcp.json in a project:

{
  "mcpServers": {
    "assistanthub": {
      "url": "https://rmassistanthub.io/mcp"
    }
  }
}

Codex

~/.codex/config.toml, through the same bridge:

[mcp_servers.assistanthub]
command = "npx"
args = ["-y", "mcp-remote", "https://rmassistanthub.io/mcp"]

Claude Code

claude mcp add --transport http assistanthub https://rmassistanthub.io/mcp

Any client, raw

curl -X POST https://rmassistanthub.io/mcp \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"get_setups_track_record","arguments":{}}}'

A real get_setups_track_record answer

The live answer as of 17 Oct 2026, 17:50 UTC, cut to the 1h window to fit here. The real one has all five windows.

{
  "asOf": 1792259421618,
  "method": {
    "call": "Each published card, treated as a buy.",
    "entry": "Close of the last candle closed when the board was built.",
    "exit": "Close of the last candle closed when the window's horizon ran out.",
    "range": "Worst and best price in between (mae / mfe), from entry.",
    "benchmark": "BTC over the same span; excess is the call's return minus it.",
    "hit": "Closed above entry.",
    "source": "1-minute exchange candles; 1y uses daily closes.",
    "minSample": 30,
    "overlap": "1d boards are sampled every 4 hours against a 24-hour horizon, so neighbouring 1d calls overlap in time.",
    "board": "One board per slot. Before 2026-10-14 a restarted server could file a second, partial board in a slot that already had one; those calls are kept but not counted here or listed in the export."
  },
  "windows": [
    {
      "window": "1h",
      "horizon": "1 hour",
      "sampledEvery": "hour",
      "top3": null,
      "resolved": 3674,
      "hits": 1756,
      "meanReturnPct": 0.05,
      "medianReturnPct": 0,
      "meanExcessPct": 0.05,
      "meanMaePct": -0.97,
      "open": 8,
      "unresolvable": 126,
      "since": 1790547501910,
      "hitRatePct": 47.8,
      "verdict": "48% of 3674 graded 1h calls closed above entry, 0.05% ahead of BTC on average.",
      "byRank": {
        "top3": {
          "window": "1h",
          "top3": true,
          "resolved": 1379,
          "hits": 637,
          "meanReturnPct": 0.04,
          "medianReturnPct": -0.06,
          "meanExcessPct": 0.05,
          "meanMaePct": -1.11,
          "open": 3,
          "unresolvable": 46,
          "since": 1790547501910,
          "hitRatePct": 46.2
        },
        "rest": {
          "window": "1h",
          "top3": false,
          "resolved": 2295,
          "hits": 1119,
          "meanReturnPct": 0.05,
          "medianReturnPct": 0,
          "meanExcessPct": 0.06,
          "meanMaePct": -0.89,
          "open": 5,
          "unresolvable": 80,
          "since": 1790547501910,
          "hitRatePct": 48.8
        }
      }
    }
  ]
}

Before an agent leans on it

• The ranking is deterministic. Each score is a weighted sum of named inputs, which every setup carries. No language model touches the score.
• The record is graded in public. The board is currently below a coin flip: on every window with at least 30 graded calls (1m, 1h, 1d and 1w), fewer than half the calls closed above entry. Ranks 1–3 hit less often than ranks 4–8 on 1h, 1d and 1w, so the top of the board is not yet doing better than the rest. They still lead on mean excess over BTC on 1w (+6.07% against +2.13%), which a few large moves can carry. On 1m, ranks 1–3 hit more often than ranks 4–8. The full numbers are on the track record.
• Trading skills are paper results. A skill installs for paper trading, and its card re-runs frozen rules on stored hourly bars. No figure on it is a filled trade.