MCP Server Setup #

sfs-mcp exposes Scottfree Sports data as Model Context Protocol tools for AI assistants. Premium subscribers can ask an assistant for model research, current odds, line movement, historical results, summaries, public consensus, weather, scores, injuries, rosters, and prediction-market context.

The assistant supplies the reasoning and writing. Scottfree Sports supplies the data tools.

Access #

If you need Premium, view pricing.

Install #

The MCP server ships with the Python CLI package.

Requires Python 3.11 or newer:

python3 --version
python3 -m pip install --upgrade scottfree-sports-cli

Verify:

sfs --version
which sfs-mcp

Your Scottfree Sports API key should look like:

sk_alphapysports_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Find it on the Account page:

https://sports-app.scottfreellc.com/app/account

What The MCP Server Can Do #

The server exposes 16 public customer tools.

Sports:

mlb, nba, nfl, nhl, ncaaf, ncaab

Model types:

over_under, won_on_points, won_on_spread

For moneyline and spread, probabilities are home-team relative. For over/under, probabilities are over-relative.

Tool Call Examples #

Your assistant calls tools for you. These examples show the exact argument shape each public tool expects.

get_daily_brief({"sport":"mlb"})

get_predictions({"sport":"mlb","model_type":"won_on_points"})

get_results({"sport":"mlb","model_type":"won_on_points","limit":25})

get_summary({"sport":"mlb","model_type":"won_on_points"})

get_odds({"sport":"mlb"})

get_sports({})

get_account_info({})

get_usage({})

get_kalshi_markets({"sport":"nfl"})

get_polymarket_odds({"sport":"nba"})

compare_market_odds({"sport":"mlb","model_type":"won_on_points"})

get_game_weather({"sport":"mlb"})

get_live_scores({"sport":"nba","date":"2026-05-23"})

get_injuries({"sport":"nba"})

get_players({"sport":"nba","search":"James","per_page":10})

get_betting_consensus({"sport":"mlb"})

Setup - Claude Desktop #

Claude Desktop runs local command MCP servers well.

1. Open Claude Desktop.
2. Go to Settings, Developer, Edit Config.
3. Add this server config:

{
  "mcpServers": {
    "scottfree-sports": {
      "command": "sfs-mcp",
      "env": {
        "SFS_API_KEY": "sk_alphapysports_PASTE_YOUR_KEY",
        "SFS_API_URL": "https://sports-api.scottfreellc.com"
      }
    }
  }
}

4. Fully quit and restart Claude Desktop.
5. Start a new chat and ask: "List the Scottfree Sports tools you can use."

If Claude Desktop cannot find sfs-mcp, use the full path:

which sfs-mcp

Then set:

"command": "/Users/you/.local/bin/sfs-mcp"

Config file locations:

Setup - Claude Code #

claude mcp add scottfree-sports sfs-mcp \
  --env SFS_API_KEY=sk_alphapysports_PASTE_YOUR_KEY \
  --env SFS_API_URL=https://sports-api.scottfreellc.com

Verify:

claude mcp list

Then start a new Claude Code session and ask:

Show me today's MLB moneyline model research and current odds.

Setup - Cursor Or Other Local MCP Hosts #

Use the same local command pattern:

{
  "mcpServers": {
    "scottfree-sports": {
      "command": "sfs-mcp",
      "env": {
        "SFS_API_KEY": "sk_alphapysports_PASTE_YOUR_KEY",
        "SFS_API_URL": "https://sports-api.scottfreellc.com"
      }
    }
  }
}

Some hosts require args instead of a single command. If so, use:

{
  "command": "python3",
  "args": ["-m", "sflow_cli.mcp_server"],
  "env": {
    "SFS_API_KEY": "sk_alphapysports_PASTE_YOUR_KEY",
    "SFS_API_URL": "https://sports-api.scottfreellc.com"
  }
}

Setup - ChatGPT #

ChatGPT custom MCP support is managed through OpenAI's app/connector flow and may require developer mode, workspace approval, or a remote MCP server. OpenAI's current help docs state that local MCP servers are not directly connected in ChatGPT; remote MCP or a supported tunnel/app flow is required.

Use Scottfree Sports with ChatGPT only if your ChatGPT plan and workspace support custom MCP apps/connectors. See OpenAI's docs:

• Apps in ChatGPT
• Developer mode and MCP apps in ChatGPT

For most customers today, Claude Desktop or Claude Code is the fastest path for local sfs-mcp.

Optional BALLDONTLIE Key #

The core Scottfree Sports tools do not require BALLDONTLIE. These tools do:

Configure:

sfs config set --balldontlie-api-key YOUR_BALLDONTLIE_KEY --use-keyring

If no BALLDONTLIE key is configured, those tools return a clear BALLDONTLIE API key not configured message instead of failing silently.

Tool Details #

get_daily_brief #

Parameters:

Returns a structured daily model brief across over/under and spread markets. Includes model pick labels, model probability, market-implied probability, signed Delta(Model-Implied), and model-position counts.

Good prompts:

Show me today's MLB daily brief. Sort the biggest model-vs-implied deltas first.

For tonight's NBA games, summarize where the model disagrees most with the market and explain what other context I should check.

get_predictions #

Parameters:

Returns current slate prediction rows and line fields.

Good prompts:

Get MLB moneyline predictions and explain the biggest difference between BLEND and IMPLIED.

Get NHL over/under predictions. For each game, show the over probability, implied probability, and current total.

get_results #

Parameters:

Returns recent graded result rows. Useful for checking whether a model has been running hot or cold recently.

get_summary #

Parameters:

Returns the summary table for all model columns, including win percentage and recent windows when available.

get_odds #

Parameters:

Returns current odds and line fields for the sport.

get_sports #

No parameters. Returns supported sports and model types.

get_account_info #

No parameters. Returns current account, plan, subscription status, and monthly limit.

get_usage #

No parameters. Returns monthly request count, limit, remaining requests, tier, and reset date.

get_kalshi_markets #

Parameters:

Returns Kalshi prediction-market data when available.

get_polymarket_odds #

Parameters:

Returns compact Polymarket event and market data.

compare_market_odds #

Parameters:

Returns Scottfree predictions, Kalshi markets, and Polymarket markets in one payload.

get_game_weather #

Parameters:

Best for outdoor sports: mlb, nfl, ncaaf.

get_live_scores #

Parameters:

Requires BALLDONTLIE key.

get_injuries #

Parameters:

Requires BALLDONTLIE GOAT tier.

get_players #

Parameters:

Requires BALLDONTLIE key.

get_betting_consensus #

Parameters:

Returns public betting consensus from Covers.

Example Premium Session #

Try this sequence in Claude Desktop or Claude Code:

You are helping me research today's MLB slate. Use Scottfree Sports tools only as data sources. Do not invent missing data.

1. Call get_daily_brief for MLB.
2. Call get_predictions for MLB moneyline.
3. Call get_odds for MLB.
4. Call get_summary for MLB moneyline.
5. If useful, call get_betting_consensus and get_game_weather.
6. Present a table with game time, matchup, current line, BLEND probability, IMPLIED probability, Delta(Model-Implied), and any notable context.
7. Separate model disagreement from betting recommendation. Tell me what is data-backed and what is uncertain.

For multi-tool examples, see MCP Workflows. For a longer walkthrough, see the Premium MCP Demo.

Troubleshooting #

sfs-mcp: command not found #

Install or upgrade:

python3 -m pip install --upgrade scottfree-sports-cli

Then locate it:

which sfs-mcp

Use the full path in your MCP config if your assistant cannot find it.

The server starts but every tool returns 401 #

The API key is missing, revoked, expired, or copied incorrectly. Verify in a terminal:

SFS_API_KEY=sk_alphapysports_PASTE_YOUR_KEY sfs account info

The server starts but tools return 403 #

Your key is valid but the subscription tier does not allow MCP. MCP is Premium-only.

Claude Desktop does not show the tools #

Fully quit and restart Claude Desktop. On macOS, use Cmd+Q; closing the window is not enough.

Scores or injuries fail #

Core Scottfree tools should still work. Scores, players, and injuries use BALLDONTLIE and need your own BALLDONTLIE key. Injuries require their GOAT tier.

Answers feel stale #

Predictions refresh after scheduled pipeline runs. Ask the assistant to call get_predictions, then inspect the predictions_date field.

See Also #

• CLI Reference
• API Reference
• Premium MCP Demo
• Account Page