gameLLM — Game Guide Assistant

A Claude Code knowledge base for video games. Scrapes guides from multiple portals, stores structured game data, and exposes everything through an MCP server and a set of Claude Code skills so you can ask questions in natural language.

Currently supports Clair Obscur: Expedition 33 out of the box. Structured data (pictos, weapons, bosses, skills, side areas, status effects, etc.) is included in the repository and works immediately without scraping anything.

Disclaimer

This project is a personal, non-commercial tool for private study and reference.

• Content ownership: All scraped content belongs to the respective source sites (Vandal, Eliteguias, Game8, IGN, Mein-MMO, Maxroll, etc.). This repository does not include any scraped data — the output/<game>/<portal>/ directories are excluded from version control and must be generated locally.
• Game intellectual property: Game names, characters, mechanics, and all related assets are the property of their respective owners (e.g. Sandfall Interactive for Clair Obscur: Expedition 33). This project is not affiliated with, endorsed by, or connected to any game developer or publisher.
• No affiliation: This project has no affiliation with any of the scraped websites or their owners.
• Terms of Service: Web scraping may be restricted by the Terms of Service of some sites. Anyone running this tool is solely responsible for ensuring compliance with the terms of any site they scrape.
• No warranty: Scraped content may be incomplete, outdated, or inaccurate. This tool is provided as-is, with no guarantees regarding the correctness or completeness of the data it produces.

---

Quick Start

git clone <repo>
cd gameLLM
pip install -r requirements.txt

Open the project in Claude Code. The MCP server (mcps/ex33mcp/) starts automatically via .mcp.json — no manual setup needed.

---

Querying Expedition 33

Use the /ex33-guide skill in Claude Code to ask anything about the game:

/ex33-guide what's the best build for Maelle?
/ex33-guide where do I find the Lost Gestrals?
/ex33-guide what pictos work well with a burn build?
/ex33-guide which bosses are in act II and what level should I be?
/ex33-guide what does the Foretell status effect do?

The skill queries the MCP server first for structured data, then falls back to scraped guides if needed.

What data is available without scraping

The output/expedition-33/data/ directory is included in the repository and covers:

File	Content
pictos.md	All pictos — names (EN/ES), effect, LP cost, location
weapons.md	All weapons — stats, scaling, passives, per character
skills.md	All skills — names (EN/ES), AP cost, character, effect
boss-levels.md	All bosses — recommended level, location, type
side-areas.md	Optional areas — act, recommended level, rewards
skill-damage-scaling.md	Damage scaling numbers per character and skill
monoco-skills.md	Monoco learnable skills and source enemies
status.md	Status effects — names (EN/ES)
zone-mapping.md	Zone name translations EN↔ES
quests.md	Side quests
expedition-journals.md	Collectible journals
music-records.md	Collectible music records
lost-gestrals.md	Lost Gestral locations
mimes.md	Mime locations
petanks.md	Petank locations
paint-cages.md	Paint cage locations
outfits.md	Outfits
haircuts.md	Haircuts
tint-upgrades.md	Tint upgrades
endless-tower.md	Endless Tower floors

MCP tools

The MCP server exposes these queryable tools (used automatically by /ex33-guide):

Tool	Description
search_pictos	Search pictos by name (EN/ES), effect, or attribute
search_weapons	Search weapons by name, passive, damage type, or character
search_bosses	Search bosses by name or location; filter by type and level range
search_side_areas	Search optional areas by zone name or reward; filter by act
get_skill_scaling	Damage scaling data for a character's skills
evaluate_picto_lumina	Recommend whether to use a picto as Picto or Lumina
search_status	Translate status effect names EN↔ES

---

Adding Content

Scrape more portals for Expedition 33

python scrape.py <guide_url> --game expedition-33

After scraping, run /ex33-index in Claude Code to update the portal index.

Add a new game

# 1. If the site isn't supported yet, generate a scraper:
/scraper-creator   ← provide the URL in Claude Code

# 2. Scrape the guides:
python scrape.py <url> --game <slug>

# 3. Bootstrap the game (generates _index.json + guide skill):
/game-setup <slug>

# 4. Start querying:
/<slug>-guide

---

Output Structure

output/
└── <game-slug>/
    ├── <portal>/
    │   ├── 001-section-title.md
    │   ├── 002-another-section.md
    │   ├── _guide.md         ← all sections concatenated
    │   └── README.md         ← section index
    ├── data/                 ← structured data (included in repo)
    └── _index.json           ← portal index

---

Technical Reference

Supported scraping sites

Domain	Scraper	Notes
vandal.elespanol.com	VandalScraper	SSR, requests
www.eliteguias.com	EliteGuiasScraper	SSR, requests
game8.co	Game8Scraper	SSR, requests
mein-mmo.de	MeinMMOScraper	SSR, requests
www.ign.com	IGNScraper	Playwright required — React-rendered
maxroll.gg	MaxRollScraper	requests
Any other URL	SinglePageScraper	Use --single-page; auto-upgrades to Playwright if JS-rendered

IGN scraper additionally requires:

pip install playwright && playwright install chromium

Scraper architecture — scrapers/base.py

GuiaScraper is an abstract base class. Subclasses implement:

• extract_guide_links(soup, guide_url) → list[{title, url}] — discovers all sections
• extract_content(soup) → Tag — extracts and cleans article content
• extract_guide_title(soup) → str — page title (optional override)
• output_slug(guide_url) → str — portal directory name

Attribute	Default	Description
REQUEST_DELAY	1.5	Seconds between requests
RECURSIVE_LINKS	False	Enable BFS link discovery

run() fetches the index, iterates sections writing numbered .md files, resumes from existing files on re-run, and writes _guide.md + README.md at the end.

BFS link discovery (RECURSIVE_LINKS = True): Some sites (IGN) only link sub-pages from category pages, not the index. BFS discovers links as it scrapes. Cached pages are not re-fetched for new links — delete the portal directory to start fresh.

Scraper notes

IGN — React-rendered; uses Playwright. Waits for .wiki-html. RECURSIVE_LINKS = True because category pages hold the individual boss/quest sub-pages. Known working selector: .wiki-html (.wiki-section fires too early).

Game8 — SSR. Content in .archive-style-wrapper. Do not use main — on game8 it only contains the membership modal.

SinglePageScraper — Tries requests first; auto-upgrades to Playwright if body < 300 chars. Steam guides work via plain requests (#profileBlock).

Adding a new scraper manually

1. Create scrapers/<name>.py subclassing GuiaScraper
2. Implement extract_guide_links, extract_content, output_slug
3. Register the domain in scrape.py's SCRAPERS dict

Use /scraper-creator in Claude Code to automate this — it fetches the target page, analyzes the HTML structure, and generates the scraper class.