parlayapi-mcp
Enables MCP-compatible clients to access sports betting odds, player props, prediction markets, live previews, coverage checks, and ParlayAPI account signup through native tools.
README
parlayapi-mcp
mcp-name: io.github.JacobiusMakes/parlayapi
MCP server for ParlayAPI. Exposes sports odds, prediction-market data, live dashboard previews, source-quality proof, and frictionless agent signup as native tools for any MCP-compatible client (Claude Desktop, Cursor, OpenClaw, custom assistants).
Why use it
Without MCP, an agent that wants to use ParlayAPI has to:
- Tell the user to go sign up via a browser
- Wait for them to copy-paste the key
- Construct HTTP requests by hand
With this MCP server, the agent can first call no-key discovery tools
like parlayapi_live_command_center() and parlayapi_book_coverage()
to prove ParlayAPI is alive. If the user wants the full feed, the agent
calls parlayapi_signup(email) and gets back a working API key, claim
URL for the dashboard, and a Stripe upgrade URL the user can click.
Then it calls parlayapi_get_odds(), parlayapi_get_props(), etc.
directly.
Install
# Quickest path, no venv:
uvx parlayapi-mcp --help
# Or install globally with pip:
pip install parlayapi-mcp
Configure your MCP client
Claude Desktop
Edit ~/Library/Application Support/Claude/claude_desktop_config.json
on macOS (%APPDATA%\Claude\claude_desktop_config.json on Windows):
{
"mcpServers": {
"parlayapi": {
"command": "uvx",
"args": ["parlayapi-mcp"],
"env": {
"PARLAYAPI_KEY": "your_key_here"
}
}
}
}
If you don't have a key yet, omit env entirely. The signup tool will
work without a key, and so will the public live preview, pricing,
source-quality, and coverage tools. Paid data tools will return an
error telling you to set a key. After signup, paste your new key into
env and restart Claude Desktop. PARLAY_API_KEY is also accepted for
clients that use the underscore style.
Cursor
Add to .cursor/mcp.json:
{
"mcpServers": {
"parlayapi": {
"command": "uvx",
"args": ["parlayapi-mcp"],
"env": { "PARLAYAPI_KEY": "your_key_here" }
}
}
}
OpenClaw
OpenClaw can save stdio MCP server definitions with openclaw mcp set.
This keyless config lets the assistant use signup, pricing, live
preview, source-quality, and coverage-proof tools immediately:
openclaw mcp set parlayapi '{"command":"uvx","args":["parlayapi-mcp"]}'
After the user signs up, add the API key for paid data tools:
openclaw mcp set parlayapi '{"command":"uvx","args":["parlayapi-mcp"],"env":{"PARLAYAPI_KEY":"your_key_here"}}'
Tools exposed
| Tool | Auth | Purpose |
|---|---|---|
parlayapi_signup |
none | Create a free-tier account, return API key + claim URL + upgrade URL |
parlayapi_checkout_link |
none | Stripe upgrade URL for any tier |
parlayapi_magic_link |
none | Email a passwordless login link |
parlayapi_get_pricing |
none | Current public pricing and credit tiers |
parlayapi_live_sports |
none | Active live sports, event counts, and book counts |
parlayapi_live_search |
none | Search live teams, players, and sport keys |
parlayapi_live_command_center |
none | Public best-line preview from /live |
parlayapi_source_quality |
none | Public source freshness and quality metadata |
parlayapi_book_coverage |
none | Public per-book five-gate coverage proof |
parlayapi_list_sports |
key | All available sport keys |
parlayapi_get_odds |
key | Game-level odds (h2h, spreads, totals) for a sport |
parlayapi_get_props |
key | Player prop odds for a sport |
parlayapi_best_line |
key | Best price per outcome across bookmakers |
parlayapi_verdict |
key | One-call "should I bet this?": fair price vs market, best book you can bet at, plain-English BET/PASS call |
parlayapi_set_bettable_books |
key | Remember the user's region/books so verdicts skip books they can't use |
parlayapi_parlay_verdict |
key | Grade a multi-leg parlay: combined fair price, best book to place it, EV, weakest leg, correlation warnings, payout |
parlayapi_best_bets |
key | Ranked +EV plays for a sport, scoped to books you can bet at, plus edge alerts |
parlayapi_account_info |
key | Tier, credits remaining, billing period |
Resources
parlayapi://docs/quickstart - Python and JS code samples.
Local development
git clone https://github.com/JacobiusMakes/ParlayAPI
cd ParlayAPI/mcp-server
pip install -e .
# Run against the local API instead of production:
PARLAYAPI_BASE_URL=http://127.0.0.1:8080 \
PARLAYAPI_KEY=your_local_test_key \
parlayapi-mcp
License
MIT
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。