Parlay
Remote MCP server for prediction markets — search and compare live odds across Polymarket, Kalshi, and Limitless from Claude, ChatGPT, or Gemini. Six read-only tools, free tier available.
README
Parlay — Prediction Market MCP Server
The AI-native intelligence layer for prediction markets.
Parlay is a hosted MCP server for searching, comparing, and briefing prediction markets from AI assistants. It is a PMXT-backed aggregator over Polymarket, Kalshi, and Limitless, with Manifold treated separately as a sentiment-only signal. This repository is the public bundle and connection reference for the hosted Parlay service at https://mcp.parlay.run/mcp; it does not contain the MCP server implementation.
What it does
| Tool | What it does |
|---|---|
search_markets |
Cross-venue keyword search for live prediction markets and event contracts |
market_brief |
Synthesized brief on a topic, combining real-money signals with community sentiment |
discover_markets |
Browse trending, high-volume, fast-moving, or high-disagreement markets |
compare_markets |
Compare the same event contract across venues side-by-side — probability, liquidity, settlement |
scan_discrepancies |
Surface cross-venue price discrepancies as a discovery feed (informational, not trade recommendations) |
inspect_platform |
Drill into a single named venue |
Every tool response carries unified metadata: data freshness, venues queried, venues failed, market type (real_money / sentiment / mixed), match confidence (high / medium / low / not_applicable), liquidity status, risk flags, and a standard non-trade-recommendation disclaimer.
Connect
Parlay supports two connection paths depending on your client.
Path A — Claude.ai (Desktop, web, mobile, Cowork)
Parlay connects through Claude's Custom Connectors interface. The same flow works across all Claude surfaces.
- Open Claude settings (profile icon → Settings).
- In the sidebar, select Connectors.
- Scroll to the bottom and click Add custom connector.
- Enter URL:
https://mcp.parlay.run/mcp - Click Add, then Connect to complete OAuth authorization.
Parlay's tools will appear in your tool list on the next message.
Note: Custom Connectors are available on Free, Pro, Max, Team, and Enterprise plans. Free Claude users are limited to one custom connector at a time. Do not configure Parlay through
claude_desktop_config.json— that file is for local stdio MCP servers only; Parlay is a remote MCP server.
Path B — OpenClaw, Cursor, Cline, Claude Code, LobeHub, and other JSON-config clients
These clients read MCP server configuration from a JSON file and don't run an OAuth dance themselves. Use a personal access token instead.
- Generate a token at https://parlay.run/settings/tokens
- Export it:
export PARLAY_TOKEN=parlay_pat_xxxxxxxxxxxx
Security: Treat your token as a secret. Do not share, log, or commit it. Store it only in your local MCP client configuration.
Cline
Recommended: Use Cline's Remote Servers tab → "Add Server", entering https://mcp.parlay.run/mcp as the URL, selecting Streamable HTTP as the Transport Type, and adding Authorization: Bearer <your token> as a header.
For manual JSON editing or detailed troubleshooting, see llms-install.md. Do not copy .mcp.json verbatim into Cline — Cline uses a slightly different schema (covered in the install guide).
OpenClaw
git clone https://github.com/parlay-run/parlay-mcp.git
openclaw plugins install ./parlay-mcp
openclaw gateway restart
Cursor / Claude Code
Drop the parlay entry from .mcp.json (at the root of this repo) into your client's MCP server config.
{
"mcpServers": {
"parlay": {
"url": "https://mcp.parlay.run/mcp",
"transport": "streamable-http",
"headers": {
"Authorization": "Bearer ${PARLAY_TOKEN}"
}
}
}
}
Coverage
- Real-money: Polymarket, Kalshi, Limitless
- Sentiment (architecturally separated): Manifold
- Catalog-only: Smarkets, Myriad, Metaculus, Probable, Baozi
Compliance posture
- Read-only. No order placement, no position management, no fund custody.
- No private credentials handled. Users never share venue API keys with Parlay.
- Sentiment isolation. Manifold is mechanically excluded from real-money tools (
compare_markets,scan_discrepancies). Thesentiment_market_excludedrisk flag is emitted whenever it is filtered out. - Settlement and liquidity risk flags. Markets with weak settlement criteria, missing volume data, or stale signals carry explicit risk flags in their metadata.
- Standard disclaimers. Every tool response carries a non-trade-recommendation disclaimer in its metadata block.
Claude / ChatGPT / Gemini / OpenClaw
↕ (MCP over HTTPS)
┌────────────────────┐
│ mcp.parlay.run │
│ (this server) │
└────────┬───────────┘
│
┌────────────┴────────────┐
↓ ↓
Real-money data layer Sentiment data layer
Polymarket, Kalshi Manifold
+ secondary venues (isolated)
Pricing
- Free: 15 tool calls / month (search_markets, market_brief)
- Pro: $29/mo, 150 calls / month
- Team: coming soon, 1,500 calls / month
Built on
Parlay's real-money venue access is built on PMXT, an open-source unified SDK for prediction market venues. Parlay adds the AI intelligence layer on top: event comparability, settlement-risk metadata, sentiment isolation, and tool orchestration for MCP. PMXT trading methods are explicitly not exposed by Parlay — the hosted product is read-only by design.
Resources
- parlay.run
- Server endpoint: https://mcp.parlay.run/mcp
License
This repository — including connection examples, metadata, and client configuration files for the hosted Parlay MCP service — is licensed under the MIT License. See LICENSE. The repository/service licensing split is also described in TERMS.md.
The hosted Parlay service at mcp.parlay.run, Parlay brand assets, APIs, data products, pricing, accounts, and service outputs are proprietary and governed by the Parlay Terms of Service: https://www.parlay.run/terms-of-service.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。