NoPunt API Review 2026

NoPunt runs an NFL prediction model and publishes everything on its site as JSON. The pitch is accountability: every pick is timestamped before kickoff, graded on a public ledger, and never edited or deleted. The free API exposes that ledger, the current week's picks, team analytics and player stat leaders. Three models vote on each game, and the site publishes the method and the calibration alongside the results.

It is not an odds API. The picks carry a market spread and a best live moneyline for context, but the product is the model's own number: a win probability and a confidence tier for the side it backs. Think of it as a second opinion to put next to a price feed, not a replacement for one.

NoPunt is NFL only, regular season only. The model has never made a postseason pick, so weeks 19 and up return a 404.

The docs list 18 endpoints. The JSON ones need no authentication, allow 60 requests a minute, and send open CORS headers, so a browser widget can call them directly. The free set splits into picks, analytics and plumbing:

• Picks. /api/picks.json returns the latest published slate. /api/picks/{season}/{week}.json returns any regular-season week from 2020 with the final score and a graded result joined to each pick.
• Performance. /api/stats.json carries the all-time record, hit rate by confidence tier and a per-season breakdown. /api/teams.json and /api/team/{abbr} add the pick history, EPA profile, coaching tendencies and playoff simulation odds for each of the 32 teams.
• Analytics. /api/power-rankings.json ranks teams by net EPA per play. /api/players.json returns stat leaders by position. /api/nextgen.json and /api/advanced-stats.json serve Next Gen Stats and Pro Football Reference style leaderboards. /api/game-wp.json returns a finished game's win-probability curve, or the season's most dramatic games.
• Plumbing. A health check, an RSS feed of the last 50 graded games, and the MCP server covered below.

Five endpoints sit behind the Pro subscription: fantasy projections, player prop projections and three CSV exports of the full pick history, weekly player stat lines and team EPA composites. They return a 403 without an active session.

Start with the current slate:

curl https://www.nopunt.com/api/picks.json

Each pick names the game by its stable nflverse id, the two teams, the side the model backs, a tier from S down to C, the picked side's win probability as a decimal, the projected margin, the market spread, a vig-free moneyline implied by that spread and the best live book moneyline. Kickoff times are ISO 8601 in UTC. Historical weeks add a result object with both scores, a correct flag and the flat one-unit profit at the spread-implied price.

Three field conventions matter, and the docs are blunt about them. predicted_margin and spread are both home-relative, so flip the sign when pick_side is AWAY. model_ml comes from the market's spread, not the model, despite the name; the model's own number is confidence. And result.correct is null on a tie, not false, so test for null before you count a loss.

Read slate_state before you treat a slate as upcoming. It is upcoming before the first kickoff, in_progress during the week and final after the last game, and between a week's end and Wednesday's new picks the newest slate is a completed one. The current slate is cached for five minutes and historical weeks for a day.

A picks feed is worth exactly as much as its track record, and NoPunt puts the whole thing behind one endpoint. When we called /api/stats.json on 16 October 2026 it reported about 1,700 graded picks across seven seasons, a straight-up hit rate around 66 percent overall and around 84 percent for the small S tier, with every season between 61 and 70 percent.

The by-tier object is the part to read twice. The docs note that accuracy rises with tier but profit does not: the S and A+ tiers hit most often and sit near flat at the spread-implied price, while the B and C tiers carry the profit. That is the sort of detail a picks site usually buries, and here it is in the JSON.

One caveat the docs state themselves: the profit series is priced at a vig-free moneyline derived from the market spread, which no sportsbook offers. Treat it as a comparable series, not a bankroll you could have had.

NoPunt ships a read-only MCP server at https://www.nopunt.com/api/mcp. It uses the Streamable HTTP transport with no auth and no session, and it serves the same free data as the JSON endpoints. A tools/list call returned seven tools when we tested it: the week's picks, a plain-English weekly briefing, the survivor board, the pick'em board, the confidence pool, the ledger summary by season, and a single matchup with its pick, line movement and injuries.

An agent-skills discovery document sits at /.well-known/agent-skills/index.json. The paid Edge signal, Kelly sizing and subscriber data are kept off the server, the same as everywhere else on the site.

The JSON API is free: no key, no signup, 60 requests a minute. Pro costs $19 a month for the five in-season months, or $69 for the season, with a seven-day free trial once the season is under way. Pro unlocks the fantasy and player prop projections, the three CSV exports and the Edge fields on the site. Pro is sold as a membership for bettors and pool players, and the API extras come with it.

The underlying play-by-play, schedules, rosters and injuries come from nflverse under CC BY 4.0, so anything you build on them needs the same attribution. NoPunt asks for a credit and a link back as well, and offers to link to production projects from its footer.

• Other sports. NFL only, regular season only. For college football or anything else, see the sports data API guide.
• Sportsbook odds. One spread and one best moneyline per pick, for context. Pair it with an NFL odds API for prices by book.
• Live data. The win-probability curves are historical play-by-play from nflfastR, not a live feed, and there is no WebSocket. The model's numbers are pregame.
• Schedules, rosters and standings. The picks feed carries kickoff times for the current week, and the team endpoints carry EPA rankings and playoff odds, but there is no season schedule, roster or standings endpoint.
• Box scores. Final scores appear inside graded picks. There are no box scores or play-by-play.

• NFL value-finder tools that already carry book prices and want an independent win probability to rank spreads against.
• Survivor, pick'em and confidence pool apps that want a ranked board with win probabilities, through the JSON or the MCP tools.
• Discord bots, dashboards and widgets. Open CORS, no key and five-minute caching make it a low-friction embed.
• Model builders who want a public, graded benchmark to measure their own NFL model against, week by week since 2020.
• AI agents that need an NFL picks source with a hosted MCP server and no credentials to manage.

NoPunt does one narrow thing and documents it well. The feed is free and keyless, the picks come with a probability instead of a hunch, every result back to 2020 is a GET away, and the docs flag their own gotchas before you hit them. The seven-tool MCP server is a bonus most free feeds do not have.

Use it as the model side of an NFL comparison, not as a price feed. Pull /api/picks.json, check /api/stats.json for the tier you plan to trust, and read the docs for the field conventions before you compute anything from the margins.

Last updated: 16 October 2026 · How providers get listed