What an agent sees through pmq

{ "mcpServers": { "pmq": { "command": "pmq-mcp", "env": { "PMQ_MCP_PAPER": "1" } } } }

1. find_markets(query="fed decision july")

   How many dissent at the July Fed meeting? will-no-one-dissent-the-july-fed-decision-20260616001928666 agent picks this	$164.99
   Fed Decision in July? will-the-fed-decrease-interest-rates-by-50-bps-after-the-july-2026-meeting	$310,450
   Fed Decision in July? will-the-fed-decrease-interest-rates-by-25-bps-after-the-july-2026-meeting	$310,450

   plus 9 more of the 12 markets returned; right column: 24h event volume

   Full-text market discovery across every Polymarket category. Works with zero credentials.

2. market(slug="will-no-one-dissent-the-july-fed-decision-20260616001928666")

   condition_id

   0x50ba1767…4ba17967

   token: Yes

   30048105…83432002

   token: No

   24681329…07939148

   closes

   2026-07-29 15:59 UTC

   settled winner

   none yet

   Slug resolved to the condition id and one token id per outcome. Those token ids drive the book and the trading tools. Long ids are shortened here for display, nothing else on this page is.

3. book(token_id=<Yes>)

   Best bid

   0.54

   592.75 shares showing

   0.02
   spread

   Best ask

   0.56

   21 shares showing

   min order 5 shares · tick 0.01 · last trade 0.57 · asks resting across 0 to 1.00: $50,243.75 · negative-risk event

   Served by the matching engine in real time, still no keys. The exchange minimums and tick come from the venue, not from constants in the library.

4. fak_buy(token_id=<Yes>, price_cap=0.58, usd=10)

   booked paper

   matched

   17.8571 shares

   price

   0.56 (cap was 0.58)

   cost

   10.00 USD

   taker fee

   0.308 USD

   cash left

   989.69 USD

   The simulator fills at the real best ask, never at the wished cap, caps the fill at the displayed size, refuses anything under the 5-share exchange minimum, and charges the documented taker fee formula, rate·p·(1-p)·shares. No order was sent: with PMQ_MCP_PAPER=1 none can be.

5. account_collateral()

   989.69 paper USD, started at 1,000.00

   One buy and its fee, nothing else. In live mode this tool reports the collateral the exchange actually sees, with a signature-type diagnostic when it disagrees with the chain.

What changes when real money is involved

Everything above ran keyless. Live trading must be switched on by the operator, in the server environment, out of the model's reach:

Structural rails on top: only fill-and-kill orders exist, so nothing ever rests unattended on the book; a fill is booked only from an exchange confirmation, never from optimism; any uncertain outcome routes the agent to reconciliation before it may trade that market again.

Run it yourself

pip install "pmquant[mcp]"

Then point your MCP client (Claude Desktop or Code, ChatGPT, LangChain, a bare SDK loop) at pmq-mcp with PMQ_MCP_PAPER=1, as in the snippet at the top, and ask the agent to find a market and trade it. Books move, so your numbers will differ from this captured session.

Every figure on this page comes from the captured session of 2026-07-04, unedited. The page itself is one file of HTML and CSS: no JavaScript, no network requests, view source to check.