imho (Python client for imho.run)

A small typed client for the imho.run API for AI assistants: Steam games like any game you name, recommendations from several games with filters, facts about one game, what is trending and newly released on Steam, title search, and identifying a game from a description. It wraps the public REST endpoints under https://imho.run/api/agent/ and the MCP endpoint https://imho.run/mcp. The API is free, read-only and needs no key.

pip install imho

Requires Python 3.9+ and httpx. Source, issues and the MCP setup for assistants: https://github.com/0x216/imho-mcp.

Usage

from imho import ImhoClient

with ImhoClient() as imho:
    picks = imho.games_like("Hollow Knight", n=3)
    for game in picks["results"]:
        print(game["rank"], game["name"], "-", game["why"], game["price"]["text"])

1 Hollow Knight: Silksong - Also Metroidvania and Souls-like, like Hollow Knight. 19.99 USD
2 Ori and the Blind Forest: Definitive Edition - Also Metroidvania, like Hollow Knight. 4.99 USD
3 Nine Sols - Metroidvania with Sekiro-style parry combat and Taoist myth. 14.99 USD

(Output from 2026-10-04. Rankings and prices change.)

games_like

imho.games_like(
    "Stardew Valley",       # name (typos, Russian titles OK), Steam appid or Steam store URL
    n=10,                   # 1..20
    lang="en",              # "en" or "ru": language of the `why` lines
    free=False,             # only free-to-play games
    coop=False,             # True = any co-op, "online" or "local" (same screen / split screen)
    steam_deck=None,        # "verified" or "playable" (playable or better)
)

Returns a dict with seed (the game that was matched), other_matches, results (each with rank, appid, name, url, steam_url, why, year, price, steam_deck, reviews, genres), list_url and attribution.

recommend

For several games, more filters, or what the user wants in their own words:

recs = imho.recommend(
    ["Stardew Valley", "Terraria"],   # 1-3 seed games
    n=10,                              # 1..24
    preferences="cozy farming, no horror",   # free text, read as Steam tags
    coop=True,                         # also "online" / "local"
    exclude=["pvp", "grind"],          # pvp, microtransactions, hard, grind, early_access, vr_only
    exclude_tags=["Anime"],            # Steam tags to leave out
    year_min=2015, year_max=None,      # release years (year_max defaults to this year)
    upcoming=False,                    # True: include unreleased games
    popularity_bias=-0.5,              # -1 more niche ... 1 more popular
    free=False, steam_deck=None, lang="en",
)
for game in recs["results"]:
    print(game["name"], "-", game["why"])
recs["preferences"]["prefer_tags"], recs["preferences"]["avoid_tags"]
# (['Farming Sim'], ['Horror'])

To refine after the user reacts, call again with liked=[...] (fills free seed slots, 3 seeds in all) and disliked=[...] (never recommended again).

trending, new_releases, search_games

imho.trending(kind="rising", n=10)      # or kind="breakouts" for new games taking off
imho.new_releases(n=10)                 # well-rated releases of the last 30 days
imho.new_releases(upcoming=True, coop=True)
imho.search_games("hollow kn", n=5)     # appid, year and links per match

Trending picks carry reviews_week and reviews_growth_pct; new releases carry release_date. While trending data is still being collected, trending() returns status: "collecting" and an empty results.

game_facts

facts = imho.game_facts("Hades")["game"]
facts["summary"]["length"], facts["summary"]["difficulty"], facts["steam_deck"]
# ('massive', 'challenging', 'verified')

Year, developers, genres, top tags, price, Steam Deck status, review numbers, Steam's short description and, when imho.run has one, a summary mined from player reviews (difficulty, length, session shape, co-op, hooks, dealbreakers).

find_game_by_description

hit = imho.find_game_by_description(
    "you play a cat in a cyberpunk city with a little drone",
    platform="pc",          # pc, playstation, xbox, nintendo, sega, mobile, browser, arcade
    year_min=None, year_max=None,
    perspective=None,       # first, third, top_down, side
)
hit["confidence"], hit["results"][0]["name"]   # ('medium', 'Stray')
hit["find_game_url"]   # imho.run/find-game with the description filled in

This one runs a language model on the server. It takes several seconds and is limited to 3 calls a minute and 20 a day per IP, so use it only when the title is unknown.

Async

import asyncio
from imho import AsyncImhoClient

async def main() -> None:
    async with AsyncImhoClient() as imho:
        facts = await imho.game_facts("1145360")
        print(facts["game"]["name"])

asyncio.run(main())

Any MCP tool

call_tool(name, arguments) calls any tool on the MCP server and returns its structured result, and list_tools() returns the tool definitions. Use them for tools added to the server after this release.

Errors

All of them subclass ImhoError, which has detail, code and status.

Limits and attribution

30 requests a minute and 1,000 a day per IP; find_game_by_description 3 a minute and 20 a day; recommend 10 new (uncached) combinations a minute and 200 a day. Responses are cached on the server.

If you show the results to people, credit imho.run ("Recommendations by imho.run") and link each game's url. Every response has an attribution field with that wording.

Development

cd python
uv run --with pytest --with pytest-asyncio --with-editable . pytest -q
IMHO_LIVE=1 uv run --with pytest --with pytest-asyncio --with-editable . pytest -q -m live   # hits the real API

License

MIT. The client is MIT-licensed; the imho.run service and its data are covered by the imho.run terms.