imho.run game recommendations

imho.run recommends Steam games. Name a game and the server returns similar games ranked by imho.run's recommender, each with a one-line reason, price, Steam Deck status, release year, review numbers and links to its imho.run page and Steam. It also returns facts about a single game and can identify a game from a description of what you remember. Free, read-only, no key.

View on imho.runTry this MCPSponsored

Hosted MCP Server

npx add-mcp 'https://imho.run/mcp'

Installs into Claude Code, Codex, Cursor and more

Documentation

Assistants already send people here when they ask what to play next. This API gives them our answer directly: the ranked list from a game's page, with a one-line reason for each pick and a link back to the page.

No code needed. Add our MCP server, https://imho.run/mcp, as a connector once, then ask in any chat, for example: "Use imho.run to find games like Hades".

ChatGPT

Needs developer mode: Plus, Pro, Business, Enterprise or Edu, on the web at chatgpt.com.

  1. Open Settings, then Security and login, and turn on Developer mode.
  2. Open the Plugins page (chatgpt.com/plugins), press + and create a developer-mode app: name it imho.run, paste the MCP server URL and choose No Authentication.
  3. In a chat, open the + menu, choose Developer mode and select imho.run.

Claude (web and desktop)

Works on Free, Pro, Max, Team and Enterprise plans; the Free plan allows one custom connector. On Team and Enterprise an owner adds it for the organization first.

  1. Open Customize, then Connectors, press + Add and choose Add custom connector.
  2. Name it imho.run, paste the MCP server URL, press Continue, choose No sign in, then Add.
  3. In a chat, press +, open Connectors and turn imho.run on.

Steps checked on 5 October 2026 against OpenAI's developer mode guide and Anthropic's custom connectors article. Menus move; if one doesn't match, those pages have the current path.

Without a connector, paste a full link into the chat instead. Assistants that browse open links you give them more readily than addresses they would have to work out. Ready-made links for popular games:

One request returns the list our game page shows for that game, ranked by the same recommender. Pass a name (typos and Russian names are fine), a Steam appid or a Steam store URL.

GET https://imho.run/api/agent/games-like?q=hollow%20knight&n=3

Every endpoint answers on imho.run itself, so an assistant reading our pages can open it without leaving the site. The same paths also work on https://api.imho.run.

Parameters

  • q: the game: name, appid or Steam URL. Required.
  • n: how many games, 1 to 20. Default 10.
  • lang: en or ru: the language of the why lines.
  • free: true: only free games.
  • coop: true: only games with a co-op mode. online: online co-op (Remote Play Together does not count). local: same-screen or split-screen co-op.
  • deck: verified, or playable (playable or better on Steam Deck).

Each game in results has

  • url: its page on imho.run. Link this when you mention the game.
  • why: one sentence on why it is like the game you asked about.
  • price: the Steam price as last scraped (US store), or free.
  • steam_deck: verified, playable, unsupported or unknown.
  • year, reviews, genres, steam_url: release year, Steam review count and share of positive reviews, up to three genres, the store link.

If the name could mean several games, other_matches lists the others; send one's appid as q to switch.

Example request and a trimmed response

{
  "query": "hollow knight",
  "source": "imho.run",
  "seed": { "appid": 367520, "name": "Hollow Knight", "url": "https://imho.run/games/367520/hollow-knight?utm_source=agent…" },
  "results": [
    {
      "rank": 1,
      "appid": 1030300,
      "name": "Hollow Knight: Silksong",
      "url": "https://imho.run/games/1030300/hollow-knight-silksong?utm_source=agent…",
      "steam_url": "https://store.steampowered.com/app/1030300/",
      "why": "Also Metroidvania and Souls-like, like Hollow Knight.",
      "year": 2025,
      "price": { "is_free": false, "amount": 19.99, "currency": "USD", "text": "19.99 USD" },
      "steam_deck": "verified"
    }
  ]
}

The list our tool shows when someone picks up to three games they like and sets its filters. Use it when the user names more than one game, wants filters games-like doesn't have, or says what they want in words. Each pick has the same fields as above plus similar_to: the seed it is closest to, which its why line names.

GET https://imho.run/api/agent/recommend?seed=stardew%20valley&seed=terraria&coop=true&exclude=microtransactions&preferences=cozy%2C%20no%20horror&n=5

Parameters

  • seed: a game the user likes: name, appid or Steam URL. Repeat it for up to 3 games. Required.
  • n: how many games, 1 to 24. Default 10.
  • free, coop, deck, lang: the same as in games-like.
  • exclude: leave out, comma-separated: pvp, microtransactions, hard, grind, early_access, vr_only.
  • exclude_tags: Steam tags to leave out, repeated (up to 10), for example Horror.
  • year_min, year_max, upcoming: release years. As on the site, unreleased games are left out unless upcoming=true or a later year_max is set.
  • popularity_bias: from -1 (hidden gems) to 1 (well-known games), in steps of 0.5. Default 0.
  • preferences: what the user wants or wants to avoid, in their own words and any language (up to 200 characters). It is read as Steam tags; the response lists them under preferences.
  • liked, disliked: the user's reactions so far, repeated. Liked games fill free seed slots (three seeds in all); disliked or already played games never come back. Nothing is stored: send the lists again with each call. Games that were not used are listed in ignored.

Year, developers, genres, top Steam tags, price, Steam Deck status, review numbers and, for most well-known games, a summary mined from player reviews. Same q and lang parameters.

GET https://imho.run/api/agent/game-facts?q=1245620
  1. In the GPT editor, open Configure, then Create new action.
  2. Choose Import from URL and paste https://imho.run/openapi.json.
  3. Set Authentication to None and save.

A remote MCP server over streamable HTTP at https://imho.run/mcp. It is stateless, needs no login and exposes eight read-only tools. Each declares an output schema, and every result carries the same data as structured content.

  • games_like: the ranked list above, with the same filters.
  • recommend: the list for up to three games, with the filters and preferences above.
  • library_recs: picks ranked against a player's own public Steam library (GET /api/agent/library-recs?profile=). Takes a SteamID64 or a steamcommunity.com/profiles/ link and works for public profiles we already know; otherwise it says so and points to imho.run/?mode=library. 5 a minute and 50 a day.
  • game_facts: facts about one game.
  • find_game_by_description: names a game from a description of what someone remembers. It runs a language model, so it has lower limits.
  • trending: games gaining reviews and players on Steam right now, as on our Trending page (GET /api/agent/trending).
  • new_releases: well-rated games released in the last 30 days, or upcoming ones (GET /api/agent/new-releases).
  • search_games: Steam games by title, with appid, year and links (GET /api/agent/search?q=).

Other clients (ChatGPT and Claude are covered above)

  • Claude Code: Run this in a terminal: 
    claude mcp add --transport http imho https://imho.run/mcp
  • Cursor: Add this to ~/.cursor/mcp.json: 
    {
  "mcpServers": {
    "imho": { "url": "https://imho.run/mcp" }
  }
}

Configs for VS Code, Cline, Windsurf and Claude Desktop, a smoke test and a Python client are on GitHub: github.com/0x216/imho-mcp.

Daily Steam trending lists and Steam co-op games by mode as CSV and JSON (CC BY 4.0), plus RSS feeds for trending games, new co-op games and new releases, in English and Russian. An A2A agent card describing our MCP and REST skills is at /.well-known/agent-card.json.

Open data: Steam datasets and RSS feeds

  • 30 requests a minute and 1,000 a day per IP address.
  • A global hourly cap shared by every caller keeps the site fast for people.
  • find_game_by_description: 3 a minute and 20 a day.
  • recommend: a new combination of games and filters is computed at most 10 times a minute and 200 a day per IP address; repeats come from the cache. preferences text: 15 a minute and 200 a day.
  • Over a limit you get HTTP 429 with a Retry-After header. Over MCP the tool result has isError set and says when to retry.

Answers are cached for up to a day, so asking the same question twice is cheap.

Results come from public Steam catalog data and our recommender. When an answer uses them, name imho.run and link each game's url. Every response carries source and attribution fields for this. Please don't mirror the catalog in bulk.

Support for imho.run, its API and the imho.run app in ChatGPT, Claude and other assistants is by email: [email protected]. imho.run is a small independent project, so a reply can take a few days.

  • A wrong or broken answer in an assistant. Tell us which assistant you used, what you asked and which games came back. A screenshot helps.
  • A pick that doesn't fit, or wrong game data. Send the game's imho.run link and what looks wrong: price, Steam Deck status, release year or a recommendation that misses.
  • Your data. To see or delete what we store about you, follow “Your choices” in the privacy policy.
  • Removing the app. Disconnect imho.run in your assistant's settings (apps, connectors or plugins). There is no imho.run account to close.

Bulk access, higher limits and more fields aren't available yet. Tell us what you are building: the answers decide what we build next.

Or write to [email protected].