Agent API & AI Skill · early access
Your trading bot and your AI assistant can see Polymarket only as it is right now. One MCP server later, they can ask what a wallet was actually worth copying, what a market's book looked like before the move, and which "opportunities" historically evaporate in a millisecond. Below: a real exchange, real numbers.
you › My bot found "swisstony" — #1 on the leaderboard, $79.8M weekly volume. Auto-copy his entries?
agent › → ticktape.copyability("swisstony")
ticktape › { copied_trades: 11,649 · win_rate: 55% · net_copy_roi_30d: −3.4% · verdict: NOT_COPYABLE — "execution-edge whale: ~43 clips per position, copier pays the spread he earns" }
agent › → ticktape.leaderboard(validated_only=true)
ticktape › [ { wallet: "S-Works", roi_30d: +11.3%, oos_month: +2.5%, grade: "B+" }, { wallet: "aenews2", roi_30d: +1.9%, oos_month: +17.2%, grade: "B" } ]
agent › Don't copy swisstony — volume isn't copyability: simulated copying lost 3.4% last month. Two wallets pass out-of-sample validation; S-Works fits your $5 sizing. Want position limits before I wire it up?
The toolbox
Shipped as an MCP server (works with Claude, and any MCP-capable runtime) plus a plain REST API — same endpoints either way. The skill layer ships with it: usage guidance, caveats and interpretation rules baked in, so the agent doesn't just get data, it gets judgment.
| Tool | What the agent learns |
|---|---|
| copyability(wallet) | simulated copy P&L for any wallet - two months, fees, slippage, out-of-sample verdict + grade |
| leaderboard(validated_only) | this week's ranked copyable wallets, with receipts and the do-not-copy red list |
| book_history(market, from, to) | full-depth order-book slices from the archive - what liquidity really looked like before the move |
| market_microstats(slug) | spread/depth/fee-curve profile of a market type - can your size even get filled? |
| arb_reality_check(market_type) | how long "guaranteed profit" moments actually last there (1 ms on crypto windows, ~65 s on slow sports) |
| trader_activity(wallet, days) | entry cadence, clip sizes, category mix - is this wallet a human or an uncopyable HFT bot? |
Quickstart
« Claude / MCP »
{
"mcpServers": {
"ticktape": {
"command": "npx",
"args": ["-y", "ticktape-mcp"],
"env": { "TICKTAPE_API_KEY": "tk_..." }
}
}
}
« plain REST »
curl -H "Authorization: Bearer tk_..." \
"https://api.ticktape.cc/v1/copyability/0xee00ba33...5cc5cea1"
This is what setup will look like. The MCP server ships to early-access users first — join the waitlist to get the install command. The MCP server and the skill definitions will be open source at launch; the API key meters the data behind them.
Why agents need this
Without history, an agent evaluates traders by leaderboard fame — the exact trap that loses money (17 of 24 top wallets were net-negative to copy).
arb_reality_check and market_microstats let an agent refuse bad ideas with numbers — "that edge lives 1 ms, we can't capture it" — before code gets written.
The skill layer encodes our methodology: out-of-sample discipline, fee curves, small-sample warnings. Your agent inherits the lessons without re-learning them on your bankroll.
Pricing
Need more volume? Write us.
Tell us what your agent should be able to ask.
MCP + REST. First API keys ship to the waitlist in order — launch pricing locked.