DEV Community

Cover image for Give your AI agent live sports prices with one MCP server
TerminalUp
TerminalUp

Posted on

Give your AI agent live sports prices with one MCP server

If you have ever asked Claude or Cursor "what's the best price on the Lions tonight" you got a shrug. Models don't carry live markets, and most odds APIs want a contract before they show you a single number.

We built a small MCP server that fixes the first half of that problem and softens the second. It exposes the TerminalUp Odds API as seven read-only tools, runs remotely with no key for sandbox data, and switches to live markets when you add a key.

What TerminalUp is

TerminalUp is a professional trading terminal for sports markets. It aggregates sportsbooks, betting exchanges and prediction markets into one workspace: compare prices across venues, route an order to the best one, manage positions from a single account. The Odds API is the same price feed the terminal runs on, published for developers and agents.

The tools

Tool What it returns
terminalup_status Service health, feed mode, coverage counts
terminalup_sports Sports and leagues currently covered
terminalup_events Events for a league, with start times
terminalup_markets Markets for an event (spread, total, moneyline, props)
terminalup_market One market: best fee-adjusted price per outcome, raw top of book, a bounded price ladder with stake, liquidity and venue counts
terminalup_find_price Natural-language lookup: "Lions -6.5" and it finds the market
terminalup_docs The API docs, so the agent can answer its own questions

Everything is read-only. There is no tool that places an order.

Install

Remote, no install, works in Claude Desktop, Claude Code, Cursor, VS Code and any client that speaks Streamable HTTP:

{
  "mcpServers": {
    "terminalup-odds": {
      "url": "https://mcp.terminalup.com/mcp"
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

Local over stdio if you prefer a process you control:

{
  "mcpServers": {
    "terminalup-odds": {
      "command": "npx",
      "args": ["-y", "terminalup-odds-mcp"]
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

With no key you get synthetic sandbox markets, labelled as such, in the exact shapes the live feed uses. To switch to live prices add a Bearer header (Authorization: Bearer tu_live_...) or set TERMINALUP_API_KEY. Keys come from terminalup.com/contact.

What it looks like in practice

Ask the agent: "Find the best available price on Detroit minus 6.5 and tell me how much I could get down at that number."

It calls terminalup_find_price, gets back the market, then reads the ladder: the best price, which venue holds it, how much stake sits at that level, and how far the price degrades as size goes up. That last part is the piece a plain odds screen never shows you. A number is only useful if there is liquidity behind it.

Why a skill as well

The same repo ships a Claude skill, terminalup-odds-api, for agents that want the raw REST, SSE or WebSocket endpoints instead of MCP. It documents keys, pagination, freshness semantics and error codes so the agent can write its own client.

npx skills add alexwooter/terminal-agent-tools --skill terminalup-odds-api
Enter fullscreen mode Exit fullscreen mode

Links

Disclosure: I'm the founder of TerminalUp. Questions and bug reports are welcome on the repo.

Top comments (1)

Collapse
 
alexshev profile image
Alex Shev •

Live prices are a good MCP use case because freshness is part of the meaning of the result, not a presentation detail. The tool response should carry observed-at time, market/source, and a clear stale-data policy so an agent cannot present yesterday’s line as live.\n\nI’d also separate retrieval from advice. Returning the quote is one capability; recommending an action based on it needs a different policy boundary and usually more context than a price feed provides.