StatMuse + Claude agents are running our fantasy team. We're undefeated.

A sneak peek at StatMuse 2.0: the StatMuse MCP pairs our sports intelligence with frontier reasoning from the model of your choice, working together as a team of agents, with you calling the shots.

In Week 4 of the StatMuse office fantasy league, our human + agent team scored 155.92 points (half-PPR). That beat the projection by 33 points and an average week in the league by 43, against the league's top-scoring team. It was the second-highest score of the season, 0.38 points off the top. The agents made the calls. A human set the rules and kept the final say.

The agents team is StatMuse + Claude. StatMuse, plugged in through the StatMuse MCP, brings the sports intelligence: who's hurt, who's getting the ball, what each defense gives up, what Vegas expects. Claude brings the frontier reasoning: it weighs the evidence, makes the moves and writes down why.

The StatMuse MCP and natural language API are now in beta for StatMuse+ members, the first of many StatMuse 2.0 launches.

Week 4 by the numbers

• Final: 155.92 to 108.32
• Record: 4-0 after inheriting an autodrafted team (no agents involved)

These are beta results. The StatMuse MCP just launched, and we're shipping new tools every week to enhance its intelligence: between Week 4 and Week 5 alone, we added pregame betting lines plus stadium and weather details for every game.

Want an agents team of your own? Get the StatMuse Fantasy Football Manager Skill on GitHub. The setup steps are at the end of this post.

How the agents team works

• Every stat comes from StatMuse. Nothing is pulled from a model's memory.
• The fantasy app's projection is the default. Every override is logged with its projected cost and the stat behind it.
• The agents run all week: three checks a day, plus one right after inactives on game days.

You choose how much to hand over

Some fantasy players will bristle at handing their team to AI. Fair enough: you don't have to hand over all of it.

In our league, we gave the agents team the keys. They set the lineup, make waiver claims and stream kickers and defenses on their own. Anything that reaches another person, like a trade or a league chat post, waits for a one-word yes.

You can set the dial anywhere:

• Hands-off: let it run the team and read the weekly log.
• Co-pilot: it researches and recommends, and you make every move.
• On demand: ask one question at a time, like "Who should I start at flex?"

Every call comes with the stat behind it, so you can overrule any of them.

Call 1: the defense it picked, then un-picked

The best call of the week was one the agents team reversed. Here's the research trail.

Friday: Buffalo or Green Bay?

Our options: Buffalo (vs New England) and Green Bay (at Tampa Bay). The app's projection had Green Bay by 0.8 points.

Asked StatMuse

• get_stats: Tampa Bay's and New England's offense, game by game
• get_stats: Buffalo's and Green Bay's defense

What came back

• Tampa Bay scored 27, 19 and 16 (20.7 a game), with 7 giveaways and 13 sacks allowed.
• New England scored 12.0 a game, with 8 giveaways and 9 sacks allowed.
• Buffalo's defense had 9 sacks, 3 takeaways and 26.0 points allowed a game. Green Bay's had 7, 2 and 30.3.

The read: In our league's scoring, holding a team to 7 to 13 points is worth 4 to a defense, and 14 to 20 is worth 1. New England's average was worth about 3 more points than Tampa's, and Buffalo had the better defense too.

The call: Start Buffalo, logged as a call against the projection (−0.8).

Saturday: the facts change

Asked StatMuse: the morning injury report and news for every player in the matchup (get_player_injuries, get_player_detail).

What came back: Baker Mayfield was out with a dislocated thumb. Undrafted rookie Jalon Daniels would make his first start.

The read: From the team log that morning:

The case for Buffalo leaned on Tampa's 20.7 points a game, but that was all with Baker Mayfield.

A rookie's first start, behind a line that had allowed 13 sacks in three games, is a good matchup for any defense.

The call: Swap Buffalo for Green Bay, back in line with the projection.

Sunday: the result

• Green Bay DEF: 8 points (3 sacks, 2 interceptions, 14 points allowed)
• Buffalo DEF: 5 points (New England scored 29)

Reversing was worth +3. When the reason behind its own call stopped being true, the agents team let it go.

Call 2: the running back who never touched the ball

When Breece Hall got hurt, Isaiah Davis became the most-added running back on the waiver wire. He was projected for 7.0 points, and the agents picked him up Friday morning as depth.

Asked StatMuse: get_stats: every Jets running back's carries and targets this season.

What came back: Isaiah Davis had 0 carries and 0 targets in three games. The backup work was going to Braelon Allen.

The read: The projection was guessing at a role the game logs said Davis didn't have.

The call: Drop Davis that afternoon for Jordan Love, insurance for Joe Burrow and the plan for Burrow's Week 6 bye. Dallas, Love's Week 6 opponent, had allowed 70.5 quarterback fantasy points in three games.

Result: Davis had 0 touches again in Week 4. Fantasy managers miss this one every week: chasing the waiver-wire name instead of checking the usage.

The common thread: StatMuse can slice anything

Both calls came down to a question the obvious stat doesn't answer. That's where StatMuse comes in: intelligence over a deep sports knowledge base, with powerful data slicing. Any stat can be cut by player, team, opponent, position, week, home or away, and even by whether a specific teammate played that game. A few from this season:

• What a defense gives up to one position, game by game: Tampa Bay has allowed 20.6, 16.1, 2.1, 19.8 and 4.8 points to tight ends.
• Games with and without a teammate: Pittsburgh's Jaylen Warren averaged 16.6 points in the two games Rico Dowdle missed, and 8.3 in the two Dowdle played.

A projection gives you one number. These slices tell the agents team when to trust it and when to go against it.

How Week 4 played out

The calls, graded:

• Kicker swap, +14.0. Will Reichard (7 for 7 on field goals, at home in a dome) replaced Jason Myers and outscored him 19.0 to 5.0.
• Brenton Strange at TE, 13.0 on an 8.0 projection. Picked up when Dallas Goedert was listed doubtful. Jake Ferguson, the other option, scored 2.6.
• The five-player stack in Jacksonville at Cincinnati: 75.0 on a 71.7 projection. With the matchup a coin flip, the agents leaned into one shootout.

The miss: Parker Washington scored 1.5 on a 14.6 projection, while Luther Burden scored 9.4 on our bench. That's −7.9, and it's on the scorecard with the rest.

It also works the waiver wire and vets trades

The waiver wire

• Weekly kicker and defense streams. For Week 5, the agents claimed Jacksonville's defense for $2 of our $100 budget: 10 sacks and 8 takeaways in four games, against a Philadelphia offense missing four starters.
• Budget discipline: under $5 spent through Week 5, with the rest saved for an injury that matters.

Trades

The agents team assumes every incoming offer is trying to win the trade, and runs it through StatMuse first. Two came in this week:

• David Montgomery and Malik Nabers for Jonathan Taylor. StatMuse had Taylor at 20.3 points a game on 21.5 carries, Montgomery at 10.2 on 11.8 (27.4 in Week 1, then 3.4, 5.8 and 4.3) and Nabers at 9.0. The agents recommended declining, and we did.
• Tyler Shough for Joe Burrow, a day later. Shough had outscored Burrow (88.3 to 79.6), but it was a buy-low on Burrow while his top two receivers were hurt. The agents recommended declining, and we did.

Our own ideas get vetted too: a Chase Brown for Travis Kelce deal is right at our bar, but with Kelce on bye in Week 5, the agents team is holding it a week.

Next season, the agents draft too

The agents didn't pick this roster. The fantasy app's autodraft did: four running backs with its first four picks, then a quarterback in round six. It took a defense in round 10 and a backup quarterback in round 12 before its first tight end in round 14, the 15th one off the board. That's why tight end is still our weak spot. The agents have been playing the hand they were dealt. Next season, they draft too, with the same research behind every pick.

Just the beginning: StatMuse 2.0

This fantasy team is the first public look at StatMuse 2.0: enhanced intelligence, a much deeper knowledge base and new capabilities. And this is just the beta. It only gets more powerful from here.

Add StatMuse to your agent

The StatMuse MCP and natural language API are in beta for StatMuse+ members. Add StatMuse to Claude, ChatGPT or any other agent that supports MCP, point it at your league on your fantasy platform of choice, and let it do the research behind every start, sit, waiver claim and trade. You make the final call.

Get the StatMuse MCP →