Muffed
About
Verified NFL stats and Sleeper league context, callable from any AI. Free, no account.
Details
- Author
- cfagan17
- Categories
- Search, Knowledge Base, Other
Jump to
Setup
Install Muffed in your MCP client (Claude Desktop, Cursor, Windsurf, and others).
Repository: https://github.com/cfagan17/muffed-mcp
Follow the installation instructions in the repository README, then restart your MCP client.
Verified NFL stats and Sleeper league context, callable from ChatGPT, Claude, Claude Code, Cursor, or any MCP-compatible client.
Free. No account, no API key, no OAuth. Read-only.
Full documentation:https://muffed.ai/mcp
An AI asked a football question today either answers from whenever its training stopped, or finds a blog post. It cannot realistically download ten seasons of play-by-play and run a rushing-yards-over-expected model.
Muffed already did that. This server makes the result callable.
Every figure comes back with the rank its source gave it and the population it was ranked against, so a claim arrives as"358.16 rushing yards over expectation, 1st of 49"rather than as a number with nothing behind it. Each response also carries a complete, quotable sentence, its own attribution, and an as-of date.
Coverage:520 verified metrics, 2016–2025 — EPA splits, CPOE, rushing yards over expectation, pressure and blitz rates, coverage splits, snap and target share, red zone and down-conversion rates, and more. Each metric reports the seasons it actually has.
Claude(web, desktop, mobile) — one click:
That opens the Add-custom-connector dialog with the name and URL already filled; you still review and confirm. By hand: Settings → Connectors → Add custom connector → pastehttps://muffed.ai/api/mcp. There is no sign-in step, because there is nothing to authorize.
Claude Code— install the plugin rather than the bare connector:
/plugin marketplace add cfagan17/muffed-mcp /plugin install muffed@muffed
The plugin ships the endpointanda skill that teaches the model how to handle what comes back: quote the pre-formed verified sentence instead of recomposing it, carry the attribution verbatim because the credit line is the licence, honour the Half-PPR and charting disclosures, never derive a rank from result ordering, and never turn verified context into a start/sit call. A bare connector gives a model the data; the plugin also gives it the discipline the data was published under. Seeskills/muffed-figures/SKILL.md.
claude mcp add --transport http muffed https://muffed.ai/api/mcp
ChatGPT— Settings → Connectors (or Developer mode) → add an MCP server → paste the same URL, no authentication.
{ "mcpServers": { "muffed": { "type": "http", "url": "https://muffed.ai/api/mcp" } } }
Every tool is read-only. Nothing writes, and no tool returns anything about a Muffed user.
1 · The panel is a reshaping, never a computation.Figures are normalized from a gated verified layer. A build gate re-derives every row back to its source coordinate and fails the build on any drift, so a source refresh that moves a number breaks CI rather than a published sentence.
2 · Verify the query template, not the sentence.A handful of parameterized queries are the templates; every instantiation inherits the gate the sources already passed. There is no arbitrary-compute endpoint and there never will be — a tool that computes a novel number on request returns a figure no gate has seen, to a model that will round it, recombine it, and attribute the result.
3 · The server ranks, filters and rounds. The model never computes.Few rows, pre-sorted, rank and population already attached. Most "the AI got the stat wrong" failures are a model doing arithmetic on returned data.
Where the source had no unambiguous rank,no rank is returned at allrather than one derived from the result set's ordering.
Rushing yards over expectation exists from2018, not 2016 — it is tracking-dependent and there is no pre-NGS version. Ask for a decade and the server returns the metric's real coverage instead of eight rows labelled as ten.
Every metric's registry entry carries the seasons it actually has, andget_metric_definitionis how a model finds that out before making a claim.
Hand it a figure your AI has already stated:
{ "claims": [{ "metric": "rb_ryoe.rush_yards_over_expected", "entity": "James Cook", "season": 2025, "value": 361 }] }
It comes backwrong, with the actual value (358.16), the source rank, and the attribution. Correct claims come backconfirmed; a metric or season Muffed does not hold comes backunverifiablewith the real coverage rather than a guess.
Structured input only — never a sentence. The caller does its own claim extraction on its own tokens; the server does a pure lookup.
No projections, no start/sit or lineup advice, no draft grades, no picks — in any tool, in any phrasing. Where a question calls for one, the tools return verified context and let you decide. That is a permanent position, not a v1 limitation.
No draft-market (ADP) data either. The sources for it are licensed to cite, not to redistribute; a page may quote a price with credit, but piping someone else's data into your AI is a different thing. Everything served here is data Muffed is licensed to build on.
No ESPN or Yahoo league support. Sleeper publishes a public read API, so loading a league needs no login. The alternative for the others would be screen-scraping or handling your credentials, and neither is going to happen.
No account, and nothing about you is stored. A league id is used to answer the request and cached briefly to be polite to Sleeper's API — never written to a database, never linked to you. A salted one-way hash of it is kept for 30 days to count roughly how many distinct leagues use the server.Sleeper usernames are not logged in any form.
Data via nflverse · NFL Next Gen Stats via nflverse · FTN Data via nflverse. League data via Sleeper's public API.
Every response carries its own attribution string verbatim — the credit line is the licence. Charting-derived figures cover the 2025 season only and are dated as such.
Muffed verifies the figures as returned, not analysis derived from them.
Issues on this repo, orhello@muffed.ai.
Individuals, use it freely. If you are calling it at volume or building it into a product, get in touch first so it stays up for everyone. The underlying nflverse data is CC-BY and free — what Muffed adds is verification, structure, freshness, and the story layer.
This repo holds the server's public manifests, the Claude Code plugin, and the documentation. The server itself runs as part of muffed.ai, which is closed-source — the plugin is a pointer to the hosted endpoint plus the skill, not a copy of the server.
.claude-plugin/marketplace.json the marketplace /plugin marketplace add reads .claude-plugin/plugin.json the plugin manifest .mcp.json the endpoint the plugin installs skills/muffed-figures/SKILL.md how to quote the figures correctly server.json MCP Registry manifest (ai.muffed/muffed) glama.json maintainer declaration for the Glama listing
Two separate listings, often confused:server.jsonis theMCP Registryentry, which is what generic MCP clients and directory crawlers read. The.claude-plugin/files are the Claude Code plugin, which is what/plugin installreads. Bumping one does not bump the other.
MCP Registry(ai.muffed/muffed) ·Glama·mcpservers.org
Search global news using natural language. Webz.io News Search API returns the most relevant articles and content, with filters for source, country, language, date, sentiment, and category.
Fetch, convert, and search AWS documentation pages, with recommendations for related content.
Search campgrounds around the world on campertunity, check availability, and provide booking links.
The Ferryhopper MCP Server exposes ferry routes, schedules and booking redirects so an AI assistant can discover connections across Europe and the Mediterranean and send users to Ferryhopper to complete bookings.
All-in-One SEO & Web Intelligence Toolkit API from FetchSERP.
MCP server that provides read-only access to HyperKitty, the web-based email archive component of Mailman 3.
At Sunrise Apps, we believe AI agents should be limitless, especially when it comes to visual data. We created ImageSorcery to bridge the critical gap in AI's ability to interact with and manipulate images directly, all while upholding the highest standards of privacy and security.
Just Domain is the domain registrar for businesses built with AI. Its remote MCP server checks availability and returns first-year and renewal pricing, plus a link to register on justdomain.ai, with DNS and WHOIS privacy in the same place. No account, no API key, read only. Endpoint: https://mcp.justdomain.ai/
Research tools, including a Sqlite-backed document stash
Semantic search over 9 free-license stock photo sources. Hosted remote server with OAuth — no API key to paste.
SerpApi MCP Server for Google and other search engine results
Sign in to leave a review
Use Google, GitHub, or an email account so ratings stay tied to real people.
No reviews posted yet.




