提交

Parlay

@parlay-run

AI-native intelligence layer for prediction markets. Parlay is a hosted MCP server that brings live prediction market data into Claude, ChatGPT, Gemini, and other MCP clients. Search and compare event contracts across Polymarket, Kalshi, and Limitless (real-money venues) with Manifold isolated as a sentiment-only signal. Six tools: search_markets, market_brief, discover_markets, compare_markets, scan_discrepancies, inspect_platform. Every response carries unified metadata — data freshness, match confidence, liquidity status, and risk flags. Read-only by design: no order placement, no fund custody, no venue API keys required. PMXT-backed real-money data layer. Free tier: 15 calls/month. Pro: $29/mo for 150 calls. Streamable HTTP transport at https://mcp.parlay.run/mcp.
概览

Parlay — Prediction Market MCP Server

MCP Badge

The AI-native intelligence layer for prediction markets.

Parlay cross-venue spread comparison across Polymarket and Kalshi

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

ToolWhat it does
search_marketsCross-venue keyword search for live prediction markets and event contracts
market_briefSynthesized brief on a topic, combining real-money signals with community sentiment
discover_marketsBrowse trending, high-volume, fast-moving, or high-disagreement markets
compare_marketsCompare the same event contract across venues side-by-side — probability, liquidity, settlement
scan_discrepanciesSurface cross-venue price discrepancies as a discovery feed (informational, not trade recommendations)
inspect_platformDrill 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.

  1. Open Claude settings (profile icon → Settings).
  2. In the sidebar, select Connectors.
  3. Scroll to the bottom and click Add custom connector.
  4. Enter URL: https://mcp.parlay.run/mcp
  5. 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.

  1. Generate a token at https://parlay.run/settings/tokens
  2. 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). The sentiment_market_excluded risk 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

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.

服务器配置

{
  "mcpServers": {
    "parlay": {
      "type": "streamable-http",
      "url": "https://mcp.parlay.run/mcp"
    }
  }
}
© 2025 MCP.so. All rights reserved.

Build with ShipAny.