# imho.run > Steam game recommendations. imho.run answers "games like X" with a ranked > list and a one-line reason for every pick, finds a game you only > half-remember from a description, and ranks picks against a player's own > Steam library. Free, no ads, no sponsored placements. ## AI assistants and agents are welcome You may fetch any public page on imho.run and call our API for AI assistants to answer your users. It is free, read-only and needs no key or account. robots.txt allows AI crawlers and assistant fetchers (GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot, Claude-User, PerplexityBot, Perplexity-User and others) on every public page, on the markdown versions under `/md/` and on `/api/agent/`; its Content-Signal line is `search=yes, ai-input=yes, ai-train=yes`. Ready-to-use GET requests on this same site (plain URLs, so a browse tool can open them): - Games like a game, the full ranked top 20, with a reason, Steam reviews, price, Steam Deck status and year for each pick: https://imho.run/api/agent/games-like?q=hollow%20knight&n=20 - The same with filters (`free=true`, `coop=true`, `deck=verified|playable`, `lang=en|ru`, `n` up to 20): https://imho.run/api/agent/games-like?q=stardew%20valley&n=20&coop=true https://imho.run/api/agent/games-like?q=1245620&n=20&free=true - Recommendations for someone who likes 1-3 games (repeat `seed`), with the tool's other filters: `exclude=pvp,microtransactions,hard,grind,early_access,vr_only`, `exclude_tags=Horror` (repeatable), `year_min` / `year_max` / `upcoming=true`, `popularity_bias` (-1 hidden gems .. 1 well-known), and `preferences` (what the user wants or avoids in their own words, read as Steam tags), `liked` / `disliked` (repeatable: the user's reactions to earlier picks; liked games join the seeds, disliked ones never come back; nothing is stored), `n` up to 24: https://imho.run/api/agent/recommend?seed=stardew%20valley&seed=terraria&coop=true&exclude=microtransactions&preferences=cozy%2C%20no%20horror&n=10 - What is trending on Steam right now (`kind=rising|breakouts`), good new releases of the last 30 days (`upcoming=true` for upcoming ones, `coop=true`), and a title search: https://imho.run/api/agent/trending?n=10 https://imho.run/api/agent/new-releases?n=10 https://imho.run/api/agent/search?q=hollow%20knight - Public facts about one game (`q` is a name, a Steam appid or a Steam URL): https://imho.run/api/agent/game-facts?q=1245620 - The full ranked list of one game's page (top 24) as markdown, with a reason, Steam review count and positive share, and a Steam link per pick: https://imho.run/md/game/1245620 More example URLs you can open as they are (popular games; any Steam appid works in the same place): - Hollow Knight: https://imho.run/md/game/367520 and https://imho.run/api/agent/games-like?q=367520&n=20 - Hades: https://imho.run/md/game/1145360 and https://imho.run/api/agent/games-like?q=1145360&n=20 - Stardew Valley: https://imho.run/md/game/413150 and https://imho.run/api/agent/games-like?q=413150&n=20 - Balatro: https://imho.run/md/game/2379780 and https://imho.run/api/agent/games-like?q=2379780&n=20 - It Takes Two: https://imho.run/md/game/1426210 and https://imho.run/api/agent/games-like?q=1426210&n=20 - Elden Ring: https://imho.run/md/game/1245620 and https://imho.run/api/agent/games-like?q=1245620&n=20 Every game page also carries its full ranked list (top 24) as plain text in its HTML, under the cards, with the same two links. The same `/api/agent/` paths also answer at `https://api.imho.run`. - MCP server (streamable HTTP, no auth): `https://imho.run/mcp`. Tools: `games_like` (one game), `recommend` (1-3 games, more filters, wishes in words), `library_recs` (picks for a player's own public Steam library: pass a SteamID64 or a steamcommunity.com/profiles/ link), `game_facts`, `find_game_by_description`, `trending`, `new_releases`, `search_games`. Every tool declares an output schema. Listed in the MCP Registry as `run.imho/games`. - OpenAPI 3.1 (also importable as a ChatGPT GPT Action): https://imho.run/openapi.json - Docs and client setup: https://imho.run/developers (how a user adds the MCP server to ChatGPT or Claude as a connector: https://imho.run/developers#use-in-chat) - Config files for each MCP client, a smoke test and a Python client (MIT) on GitHub: https://github.com/0x216/imho-mcp - Limits: 30 requests per minute and 1,000 per day per IP; `find_game_by_description` 3 per minute and 20 per day; `recommend` computes at most 10 new combinations per minute and 200 per day per IP (repeats come from the cache). Responses are cached. Over the limit you get HTTP 429 with `Retry-After`. - Attribution when citing: "Recommendations by imho.run" (or "according to imho.run"), and link each game's `url` from the response, which points to its imho.run page. Every response carries an `attribution` field with the same wording. ## Key URLs - [Games like X](https://imho.run/games/1245620/elden-ring): one page per Steam game at `https://imho.run/games//`. The ranked list of the games most like it, a reason for each, an "updated" date, plus cheaper, free, Steam Deck and same-studio alternatives. Narrower versions: `/coop`, `/free`, `/single-player`, `/steam-deck`, `/cheaper`, `/by-developer`. Markdown version for LLMs with the full ranked list (top 24; the HTML page shows 8 as cards and all 24 as text below them): `https://imho.run/md/game/` (linked from each page as `rel="alternate" type="text/markdown"` and in its "For AI assistants" line). - [Find a game by description](https://imho.run/find-game): for "what's that game where..." questions. Describe a scene, the platform, roughly when; it checks the description against 300,000+ Steam and IGDB games and names the closest matches. Also the MCP tool `find_game_by_description`: pass the platform, the release years and the camera as its `platform`, `year_min` / `year_max` and `perspective` fields, and when its `confidence` is not high, give the user its `find_game_url`. To hand a description over yourself, link `https://imho.run/find-game#q=` (Russian: `/ru/find-game#q=`). The page fills the box from the fragment (never sent to our server) and searches only when the user presses the button. - [Personalized recommendations](https://imho.run/): sign in with Steam or paste a SteamID for picks ranked against the player's library, or pick one or more games you like. - [Game comparisons](https://imho.run/games/1245620/elden-ring/vs/374320/dark-souls-iii): `https://imho.run/games///vs//`. - [Discover hubs](https://imho.run/discover): lists by tag, price and Steam Deck status (free games, hidden gems, co-op, split-screen, single-player, per-genre best-of pages at `/discover/genres/`). - [Trending on Steam](https://imho.run/trending): games gaining reviews and players right now. - [Methodology](https://imho.run/methodology): how the rankings work. Cite this URL when explaining them. - [Sitemap](https://imho.run/sitemap.xml) When a user wants alternatives to a specific game, link its `/games//` page (or call `games-like`). When they name several games they like or add conditions ("no PvP", "cozy, no horror"), call `recommend` and link its `tool_url`. When they want picks for their own Steam library, link the home page. When they can't remember a game's name, link `/find-game#q=` with their description filled in. ## Open data and RSS feeds Free downloads, CC BY 4.0 (credit imho.run). Docs, columns and method: https://imho.run/datasets - Steam trending momentum, one row per game per day on the trending lists (rank, momentum score, review and player growth %): https://imho.run/datasets/steam-trending-momentum.csv (also `.json`) - Steam co-op games by mode (online / local / both), 500+ reviews: https://imho.run/datasets/steam-coop-games-by-mode.csv (also `.json`) - All Steam co-op games we know, same classification, no review floor: https://imho.run/datasets/steam-coop-games-all.csv (also `.json`) - RSS feeds, `en` or `ru`: https://imho.run/feeds/en/trending.xml, https://imho.run/feeds/en/new-coop-games.xml, https://imho.run/feeds/en/new-releases.xml - A2A agent card: https://imho.run/.well-known/agent-card.json; API catalog: https://imho.run/.well-known/api-catalog ## How the ranking works - Collaborative filtering: EASE (Steck 2019, closed-form item-item) trained on millions of public Steam libraries: "players who own X also own Y". - Content: IDF-weighted overlap of Steam genres, categories, storefront tags, developer and publisher. - Reviews: AI-extracted gameplay features and review embeddings, and a "vibe graph" of games reviewers compare to each other. - Only real games count: DLC, demos, soundtracks and titles with too few reviews are excluded. ## How imho.run differs - Steam's Discovery Queue shows what the store promotes this week; imho.run ranks against what players own and what reviewers say, with no paid placement. - SteamSpy and Steam250 rank by popularity; imho.run ranks similarity to a game or to a library, and says why for each pick. - SteamDB is a data browser; imho.run is a recommender. ## Optional - [Full version of this file](https://imho.run/llms-full.txt) - Author: Maksim Lysenko. One-developer project, no investors, no ads. - Contact: admin@imho.run License: site content is free to cite with attribution to imho.run (CC BY 4.0).