mcp-ps-store is an MCP (Model Context Protocol) server that exposes your PlayStation Network (PSN) gaming history—playtime, trophies, purchased game library—to AI assistants, enabling them to recommend games based on real data rather than guessing. The project is written in Python and requires no official PSN API.
Purpose PSN has no public API. This server communicates with the same internal API used by the PlayStation mobile app, authenticating via an NPSSO cookie obtained from a browser session. All requests are read operations and never write anything to the account.
Main capabilities The server exposes MCP tools and provides a recommend_games prompt—a ready-made "build my profile and recommend games" scenario that accepts optional guidance such as "shorter" or "two-player co-op".
Usage Requires Python 3.11+. After installation, log in in the browser at https://www.playstation.com，打开 https://ca.account.sony.com/api/v1/ssocookie to obtain the returned {"npsso":"..."}, copy the 64-character npsso value and run the login command. The token is saved in ~/.mcp-ps-store/tokens.json with permissions 0600. The access token is valid for 1 hour, the refresh token for 10 days, but the NPSSO itself lasts about two months, so the server automatically re-authenticates when the refresh token expires, and a single psn-login can actually last about 60 days.
Connecting clients Configure it in claude_desktop_config.json. Codex communicates with local MCP servers via stdio and is also supported; it can be added via codex mcp add or by editing ~/.codex/config.toml, and checked with /mcp in a Codex session. ChatGPT cannot launch local processes; custom connectors must be added in developer mode (Settings → Apps → Advanced) and must be a public HTTPS endpoint supporting SSE or Streamable HTTP. Therefore, you need to switch the server to HTTP transport and publish it, then expose it over HTTPS http://127.0.0.1:8000/mcp（可使用 OpenAI's Secure MCP Tunnel or a tunnel such as cloudflared/ngrok), then add the generated URL as a custom connector. Note: this exposes your PSN history to anyone who finds the URL, and the server itself has no authentication, so you should put the tunnel behind authentication, run it only when needed, and prefer the Codex CLI if a local client is available.
Use cases Suitable for users who want an AI assistant to make game recommendations based on their personal real PSN gaming data (playtime, trophies, purchased library).
Limitations Playtime exists only for PS4/PS5/PC versions. PSN reports only trophies for PS3 and Vita games, so those games show hours: null. There is no PS Store catalog here. The store is entirely client-side, and its GraphQL API only accepts persisted queries, while Sony rotates the hashes on every deployment, making it impractical to keep them working. Recommendation candidates come from the model's own game knowledge, while psn_owned_games prevents suggesting games already purchased. The purchased game library is fetched via PSN's private GraphQL API; if Sony changes the query, this tool will stop working; other features continue to work, and a backlog_unavailable note is added to the summary. Playtime and trophies are joined by title, so they may be merged when a remaster and the original share a trophy set. Version, platform suffixes, and Roman numerals are intentionally normalized—Alan Wake II and Alan Wake 2 are treated as the same game. This API is unofficial, and Sony can change it at any time.
The project structure uses a layered architecture: API layer (MCP tools), core layer (services, repositories, DTOs), DI container, and infrastructure layer. Core technologies include Python 3.11+, the MCP protocol, httpx (async), dependency-injector, pydantic-settings, Poetry, Pytest, and pytest-asyncio. The license is MIT.
