Prediction Market Signal Mcp

io.github.ciinkwia由 GitHub 账号 ciinkwia 发布v0.1.0更新于 Nov 10, 2026

Polymarket + Kalshi odds, movers, gaps, watch, and X sentiment for AI agents. Pay per call via x402.

已验证STDIO仅桌面网页搜索与抓取金融

概览

AI 生成的概览

让助手获取 Polymarket 与 Kalshi 预测市场赔率、异动、历史、跨平台价差和 X 舆情,按次以 USDC 付费。

可以做什么
  • 跨 Polymarket 与 Kalshi 搜索并排序市场,获取头条赔率、跨平台价差与 7 天历史
  • 查看最大概率波动、单个市场的概率序列,以及提及某只美股代码的市场
  • 比较两平台分歧最大的问题,列出即将结算或刚刚结算的市场
  • 监控指定市场或主题,只返回自上次检查以来的变化
  • 把市场赔率与 X/Twitter 讨论的 AI 解读合并到同一次查询
  • 无需钱包即可预览一份免费示例结果
适用场景
适合让助手查询选举、美联储决议、宏观事件等现实问题的群体预测赔率,比较两个预测市场对同一问题的定价,或长期跟踪概率变化而不订阅服务。
运行要求
通过 npx 在本地运行(需 Node.js)。需要一个在 Base 上存有 USDC 的 Coinbase CDP 钱包,通过 CDP_API_KEY_ID、CDP_API_KEY_SECRET 和 CDP_WALLET_SECRET 配置;CDP_ACCOUNT_NAME 与 X402_MAX_PRICE 可选。需要访问付费端点和 Base 结算的网络。免费的 pm_example 工具无需钱包。
安装前请注意
每个付费工具都会从你自己的钱包实际支出 USDC,每次 0.01 至 0.05 美元;X402_MAX_PRICE 只限制单次价格,不限制总花费。三项 CDP 凭据(含钱包密钥)会保存在客户端配置中。查询内容和市场 ID 会发送给第三方端点,pm_sentiment 还会抓取公开的 X/Twitter 帖子。输出仅为市场数据,不构成投资建议。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Prediction Market Signal Mcp,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

prediction-market-signal-mcp

Live Polymarket + Kalshi prediction-market odds, movers, price history, cross-platform gaps, watch, and X/Twitter sentiment inside Claude, Cursor, or any MCP client — no API key, no subscription, no monthly minimum. Your agent pays a fraction of a cent per call in USDC on Base using x402.

pm_search          $0.01/call    markets matching a question, Polymarket + Kalshi, rankedpm_odds            $0.02/call    headline per platform, cross-platform gap, 7-day history, summarypm_movers          $0.02/call    biggest probability swings (Polymarket, liquid markets)pm_history         $0.02/call    one market's probability series, with first/last/high/lowpm_ticker_events   $0.01/call    open markets mentioning a US stock ticker or companypm_sentiment       $0.05/call    markets AND X/Twitter chatter on the same question, one callpm_watch           $0.02/call    only what moved since your last check (stateless cursor)pm_gaps            $0.02/call    biggest Polymarket-vs-Kalshi disagreements, same questionpm_closing         $0.01/call    markets settling soon, both platformspm_resolved        $0.01/call    markets that just settled, with the outcomepm_example         free          a real capped pm_odds result, no wallet needed

Facts only — not investment advice. Every answer is implied probability, volume, and price history from the markets themselves (plus, for pm_sentiment, a plain read of what X/Twitter is saying). Nothing here is a buy/sell/hold signal, a trade recommendation, or financial advice.

Why this exists

Prediction markets (Polymarket, Kalshi) already price real-world questions — elections, Fed decisions, macro events — as crowd-forecast odds. Reading them across two platforms usually means two different SDKs, two rate limits, and no cross-platform comparison. This server does both in one call, charges per request instead of a subscription, and there is nothing to cancel.

Tools

pm_search — $0.01

Markets on Polymarket + Kalshi matching a question or topic, ranked by relevance, each with probability, 24h change, volume, close date, URL, and id (use the id with pm_history or pm_watch). Inputs: q (required), limit (optional, 1-50, default 20).

pm_odds — $0.02

"What do the markets say" for one question — the headline market per platform, the cross-platform probability gap, 7-day history, and a one-line summary:

json
{  "headline": { "polymarket": {...}, "kalshi": {...}, "gap_pts": 4 },  "history_7d": { "points": [...] },  "summary": "..."}

Input: q (required). Use pm_search instead when you want the full ranked list.

pm_movers — $0.02

Biggest 24h/7d probability swings among liquid Polymarket markets — what's newly hot. Inputs: window (optional, 24h default or 7d), limit (optional, 1-50, default 20).

pm_history — $0.02

One market's probability series over time, first/last/high/low. Inputs: id (required — a Polymarket condition id/slug/numeric id, or a Kalshi ticker, from pm_search), interval (optional, 1d/1w/1m/max, default 1w).

pm_ticker_events — $0.01

Open Polymarket + Kalshi markets mentioning a US stock ticker or company (a bare ticker like TSLA is also matched against the company name, Tesla). Facts only — implied probability, 24h change, volume — never a trade signal. Inputs: q (required — ticker or company/topic), limit (optional, 1-50, default 20).

pm_sentiment — $0.05

What the markets say AND what X/Twitter is saying about the SAME question, one call: headline market probability, an AI read of X chatter's yes/no lean, key themes, driving accounts, notable post URLs, and an aligned / diverges / no_clear_read verdict comparing the two:

json
{  "summary": "Markets price \"...\" at 62%. X chatter (14 recent posts) leans YES — aligned with the market.",  "agreement": "aligned",  "market": { "polymarket": {...}, "kalshi": {...}, "gap_pts": 3 },  "x": { "lean": "yes", "lean_note": "...", "summary": "...", "key_themes": [...] }}

Input: q (required). No market matched on either platform = no charge.

pm_watch — $0.02

Monitor 1-20 market ids or a topic, get back only the change since your last check: current probability, change_since_cursor (probability points), a moved flag, and (for a topic watch) newly-listed markets. Pass the response's cursor back in next time — it's stateless, so you can run any number of watches in parallel. Inputs: ids (1-20 comma-separated ids) OR q (a topic) — provide one, not both; cursor (optional, omit on the first call); min_move (optional, probability points, default 3). Nothing moved and nothing new = no charge — the fresh cursor still comes back so your monitor keeps its place.

pm_gaps — $0.02

Biggest Polymarket-vs-Kalshi disagreements on the SAME question — matched by topic, close date, and shared numbers, with a match_confidence score and a polarity check so opposite-side matches never show as a misleading gap. Served from a snapshot rebuilt every 30 minutes. Inputs: limit (optional, 1-50, default 20), min_gap (optional, probability points, default 5), q (optional topic filter). No matches at that gap size = no charge; snapshot not built yet after a cold start = free, retry shortly.

pm_closing — $0.01

Markets on both platforms closing within 24h or 7d, sorted by volume. Inputs: window (optional, 24h default or 7d), limit (optional, 1-50, default 20), q (optional topic filter).

pm_resolved — $0.01

Markets on both platforms that settled within 24h or 7d, with the final outcome and last price — for grading a forecast against reality. Same inputs as pm_closing.

pm_example — free

A real pm_odds result for a fixed question ("fed rate cut"), capped to 5 markets and 12 history points, refreshed every 6 hours. No wallet, no payment, no x402 needed — use it to see the answer shape before spending on any paid tool. No inputs.

Setup

You need a Coinbase CDP wallet funded with a little USDC on Base — that wallet pays per call. Create one at portal.cdp.coinbase.com (API key + wallet secret). pm_example works without any of this — it's free.

Claude Desktop / Claude Code — add to claude_desktop_config.json (or .mcp.json for Claude Code):

json
{  "mcpServers": {    "prediction-market-signal": {      "command": "npx",      "args": ["-y", "prediction-market-signal-mcp"],      "env": {        "CDP_API_KEY_ID": "your-key-id",        "CDP_API_KEY_SECRET": "your-key-secret",        "CDP_WALLET_SECRET": "your-wallet-secret",        "X402_MAX_PRICE": "0.05"      }    }  }}

Cursor — Settings → MCP → Add new MCP server, same command/args/env as above.

X402_MAX_PRICE is a hard per-call ceiling in USD. If the endpoint ever quotes more than this, the request is refused before anything is signed — no charge. The default 0.05 already covers every paid tool above (the priciest is pm_sentiment at $0.05) — you don't need to raise it.

Payment

x402, bring-your-own-wallet. Every paid tool pays from the wallet you configure — your own Coinbase CDP account, funded with USDC on Base. There is no server-side wallet and no shared account: each call signs its own USDC authorization, and the endpoint's facilitator settles it on Base. Gas is covered by the facilitator; you spend USDC only.

Cost & safety

  • You are only ever charged for a successful answer. Invalid inputs are rejected before payment (HTTP 400, no settlement).
  • Nothing found = nothing charged. An empty pm_search / pm_ticker_events / pm_gaps / pm_closing / pm_resolved / pm_sentiment comes back as no_results with charged: false — the payment is never settled.
  • pm_watch only charges when something moved or something new appeared. A check that finds nothing new is free — poll it on a schedule without worry.
  • pm_gaps before its 30-minute background snapshot has finished its first build answers free (503) instead of charging for an incomplete answer.
  • If the upstream market data fails after payment, the API returns an explicit error and logs the call for refund rather than pretending it worked.
  • Facts only, not investment advice. Every number is a market price or a plain read of public chatter — never a recommendation to trade, a price target, or a buy/sell/hold call.

How it works

MCP client → prediction-market-signal-mcp → 402 challenge → sign USDC authorization           → facilitator settles on Base → data returned

Protocol: x402 v2 (CAIP-2 networks, PAYMENT-REQUIRED challenge header).

License

MIT

来源:README.md,提交 0072f59

工具

0
SourceWeft 尚未在沙箱中启动过该软件包,因此其工具未知。

版本历史

1
  1. v0.1.0最新Sep 28, 2026

评分

已满足适用检查项的 50%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通25 分

    请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。

  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分