TipMaster Expert Soccer Predictions
Football (soccer) predictions from players with a track record going back to 1997, for Claude, ChatGPT and any MCP client. Free, no account.
The short version. TipMaster is a free football prediction game running since 1997, with nearly 30 years of prediction-game experience. Expert groups are selected from recorded player performance. Prediction shares measure agreement, not calibrated probabilities; assess confidence using measured accuracy, its date window and sample size.
For each match you get the most likely final score, the home/draw/away split of the Top 20%, Top 25% and All players, and, once played, the result. A separate answer says how often each group was right: correct home/draw/away calls, and exact scores. Connect https://tipmaster.de/api/agent/mcp to your assistant and ask, for example, "What will Bayern vs Freiburg end, according to TipMaster's experts?"
What you can ask
| Question | Tool |
|---|---|
| "What's the most likely score of Arsenal vs Chelsea?" | get_match_prediction |
| "Which Premier League matches this weekend do the best players call differently from everyone else?" | find_matches |
| "How often are TipMaster's top players right on the Bundesliga?" | get_accuracy |
| "What did TipMaster's experts pick in this week's Classic round?" | get_expert_picks |
The two game modes behind the data
| Mode | What players predict | How the top groups are ranked |
|---|---|---|
| National | The exact score of real matches in the Bundesliga, Premier League, LaLiga, Serie A, Ligue 1 and other top leagues, 15 countries | Home/draw/away hit rate over the 8 weeks before the match's week, at least 20 settled predictions |
| Classic | Home, draw or away on 25 matches a week, since 1997 | Career points per match, at least 100 career matches |
A match is published once TipMaster players can no longer enter or change their prediction for it, so the numbers are final. World Tour, Fanclub and Friends & Family join as they launch.
One match
{
"id": "daily:4821",
"mode": "national",
"competition": "Bundesliga",
"home": "Bayern München",
"away": "SC Freiburg",
"status": "locked",
"result": null,
"scoreRecorded": true,
"mostLikelyScore": { "home": 2, "away": 0, "share": 0.21 },
"bands": {
"basis": "national_hit_rate_8_weeks_before_week",
"items": [
{ "id": "top20", "label": "Top 20%", "sample": 96, "majority": "home",
"outcome": { "home": 0.83, "draw": 0.12, "away": 0.05 },
"scorelines": [{ "home": 2, "away": 0, "share": 0.27 }, { "home": 3, "away": 1, "share": 0.18 }, { "home": 2, "away": 1, "share": 0.14 }] },
{ "id": "top25", "label": "Top 25%", "sample": 120, "majority": "home", "outcome": { "home": 0.81, "draw": 0.13, "away": 0.06 }, "scorelines": [] },
{ "id": "all", "label": "All players", "sample": 480, "majority": "home", "outcome": { "home": 0.74, "draw": 0.16, "away": 0.10 }, "scorelines": [] }
]
}
}Illustrative numbers, shortened. Shares are fractions of the group and sum to one. Classic matches carry scoreRecorded: false and no scores. A withheld group has outcome: null, sample: null and a reason.
Accuracy
| Field | Meaning |
|---|---|
majorityCorrect | Share of matches where the group's most common home/draw/away call was right. |
outcomeCorrect | Share of the group's individual predictions with the right outcome. |
exactCorrect | Share with the exact final score (National only). |
National covers the last 8 weeks; Classic covers the current season.
Endpoints
| URL | Returns |
|---|---|
POST /api/agent/mcp | MCP server, Streamable HTTP, stateless, no authentication. Tools: get_match_prediction, find_matches, get_accuracy, get_model_comparison, get_expert_picks, get_consensus, list_matches. |
GET /api/agent/v1/consensus?mode=&status= | Full per-match records, including bands and mostLikelyScore. |
GET /api/agent/v1/consensus/{id} | One match. |
GET /api/agent/v1/experts?game= | This week's Classic round of 25 matches with the expert cohort. |
GET /api/agent/v1/matches?mode=&status= | The matches without the prediction fields. |
GET /api/agent/v1/openapi.json | The OpenAPI 3.1 document. |
get_model_comparison returns an experimental model benchmark trained on stored Sportmonks results, with a later test window and same-match comparisons against the published groups. Rates are withheld below 50 matched fixtures. It is a retrospective evaluation, not a calibrated forecast or a claim that the model beats experts.
Connect
Step-by-step setup for Claude and ChatGPT, including phone availability and example questions.
Claude
Customize, Connectors, Add custom connector, URL https://tipmaster.de/api/agent/mcp. Choose No sign in if asked. Enable the connector in your conversation. Account and workspace restrictions may apply.
ChatGPT
On the web, open Plugins, select +, then Add custom MCP server. Enter https://tipmaster.de/api/agent/mcp and choose No authentication. Create and install the plugin, then select it with @ in a conversation. Account and workspace restrictions may apply.
Any MCP client
{
"mcpServers": {
"tipmaster": { "url": "https://tipmaster.de/api/agent/mcp" }
}
}curl
curl -s "https://tipmaster.de/api/agent/v1/consensus?mode=national" | jq '.matches[] | {home, away, mostLikelyScore}'Rules of use
Open access, 60 requests per minute per IP address. Responses carry a 60 second cache header, so there is no point polling faster. Please link to https://tipmaster.de/agents/ wherever you show or cite the numbers; every record carries that URL in its source field. The service is informational. It is not betting advice and it contains no odds. Use is subject to the terms of use.
FAQ
Is this betting advice?
No. TipMaster is a free prediction game with no stakes. The service reports what its players predicted and how often they were right. It carries no odds, no bookmaker data and no recommendation.
Is it free? Do I need an account or an API key?
It is free, with no account and no key. Each IP address may make 60 requests per minute, and responses can be cached for 60 seconds.
Where do the predictions come from?
From TipMaster players. TipMaster has run since 1997. Expert groups are selected from recorded player performance; the available history differs by mode. National players predict the exact score of real top-league matches; Classic players pick home, draw or away on 25 matches a week.
Who are the Top 20% and Top 25%?
National: the players with the best home/draw/away hit rate over the 8 weeks before the week of the match (at least 20 settled predictions), so a group is never picked by the result it is judged on. Classic: the round's players with at least 100 career matches, ranked by career points per match since 1997. All players includes everyone. No bottom group is published.
Can predictions still change?
No. A match appears once TipMaster players can no longer enter or change their prediction for it, so the numbers are final. If you want to predict yourself, play at https://tipmaster.de/.
Why is a group sometimes missing?
Every published share counts at least 20 players. A smaller group is withheld, and so is a Top 20% split whenever it would differ from Top 25% by fewer than 20 players, so no small group can be worked out by subtraction.
Can I see individual players' predictions?
No. Only group aggregates are published. No individual player, name or record is ever shown. Single predictions never leave the system, and players who object to data use are left out entirely.
How should I credit the data?
Link to https://tipmaster.de/agents/ wherever you show or cite the numbers. Every record carries that URL in its source field.