Summary
MyGM is a read-onlyModel Context Protocol connector. It reads public Sleeper league data via Sleeper's public API and presents it back to you and to Claude. It does not collect, store, sell, or share any personally identifiable information.
What we read
- Sleeper league rosters, matchups, transactions, drafts, and standings — sourced from the public
api.sleeper.appendpoints. - NFL player metadata and weekly stats / projections — also from Sleeper's public API.
All upstream data is already public on Sleeper. MyGM never sees your Sleeper password, account email, payment info, or any credentials.
What we cache
- League history snapshots, slim NFL player index, and weekly stats / projections are cached in Vercel Blob (
cache/...) to reduce upstream load. Cached data mirrors the public Sleeper data — nothing additional. - No per-user identifiers are stored alongside the cached data.
Logs
Vercel function logs may contain transient request IPs / user agents (standard infra logging). These rotate per Vercel platform retention. They are not joined to any user profile by MyGM.
Tracking
MyGM has no analytics scripts, no advertising pixels, no third-party trackers. The site does not set cookies for tracking.
Rate limiting
The API endpoints (/api/mcp, /api/dashboard, /api/widgets) are protected by an in-memory per-IP token-bucket rate limiter to deter casual abuse. It uses only the request IP and User-Agent for that check (the User-Agent lets known Claude/Anthropic clients bypass the limit) and does not store them or link them to any user identity. /api/chat has its own, IP-only limiter — see Howie below.
Howie (the chatbot)
Chat messages you send to Howie are relayed through the Vercel AI Gateway to a third-party model provider (e.g. Anthropic) so it can generate a reply. We do not store your conversation text server-side — the transcript lives only in your browser's sessionStorage until you close the tab or clear it yourself.
- We keep anonymous, aggregate usage counters (message counts, rate-limit blocks, grounding-check warnings) — not tied to any message content or identity.
- If you tap 👍/👎 on a reply, we store that vote, which tools ran to produce it, a message/request id, a timestamp, and a one-way hashed-IP correlator (never the raw IP) — never the words of the conversation.
- Rate limiting on
/api/chatis its own, simpler limiter (no User-Agent check): a short-lived, in-memory check keyed to your IP, never persisted or linked to your identity; anything logged uses a truncated one-way hash of the IP, never the IP itself.
Contact
Questions: support.
Last updated: 2026-07-13.