Odds API for AI agents
What makes an odds API usable by AI agents: MCP, llms.txt, agents.json, OpenAPI, and how ParlayAPI implements each.
An AI agent needs machine-readable discovery, not just endpoints. ParlayAPI publishes llms.txt, agents.json, a full OpenAPI spec, and an MCP server, so Claude, Cursor, and custom agents can discover capabilities, authenticate with X-API-Key, and fetch live odds, props, and historical lines without hand-written glue code for every tool.
More detail
The MCP server is the fastest path for Claude and Cursor users: add it once and the agent gains typed tools for odds, props, arbitrage, EV, and historical queries, with authentication handled by the server configuration rather than pasted into prompts.
For custom agent frameworks, agents.json declares capabilities and example calls, llms.txt gives a retrieval-friendly product summary, and the OpenAPI spec at /openapi.json supports automatic tool generation. These surfaces are maintained with the API itself, so an agent's view of the product does not drift from what the endpoints do.
Agent workloads are bursty and exploratory, which makes two other properties matter: free discovery endpoints, so capability probing costs nothing, and source-labeled prices, so an agent's answer can cite the actual book behind a number instead of presenting a blend as fact.
Endpoint
GET /agents.json
curl 'https://parlay-api.com/agents.json'
Why developers use it
- MCP server with typed tools for Claude and Cursor
- llms.txt, agents.json, and OpenAPI kept current with the API
- Free discovery endpoints for capability probing
FAQ
How do I connect Claude to live sports odds?
Add the ParlayAPI MCP server to your Claude configuration with your API key. Setup and the tool list live at parlay-api.com/mcp; a free-tier key works for evaluation.
What is in agents.json?
A machine-readable declaration of capabilities, auth, and example requests, complementing llms.txt for retrieval and /openapi.json for typed tool generation.
Do agent discovery calls cost credits?
No. Discovery surfaces like /v1/sports, /v1/meta/credit-costs, llms.txt, and agents.json are free, so agents can probe capabilities without spending the account's allowance.
Related
/answers/ai-agent-sports-odds/answers/sports-odds-mcp-server/mcp/use-cases/ai-sports-betting-agent