Palworld TCG public API

https://dev.palworldtcg.gg/api/v1

Every endpoint below is relative to this URL. All responses are JSON with a stable envelope: { data, meta } for lists, { data } for single items, { error, message } on failures.

The card, set, price and deck routes have no per-key limit. Please be a good citizen: cache responses locally, use ETag/If-None-Match, and don't hammer the endpoints in a tight loop. Every route ships aggressiveCache-Control headers (5 minutes for cards, 1 hour for sets and prices, 1 minute for decks) so Vercel's edge cache absorbs most traffic before it reaches our origin.

/meta is the exception: it is capped per IP per day, because its snapshot only changes daily. See below.

If we start seeing abuse, we'll flip on IP-based limits without notice. Build defensively.

This is the same aggregated data the /meta pages render. Every archetype carries decklist — its representative 50-card main deck as { slug, count } entries, the medoid of that archetype's population — alongside win rates with confidence intervals, the head-to-head matrix, play order and composition bands.

Those decklists are centroids computed over thousands of games, not anybody's saved deck: there are no per-game rows, no players and nothing that can be joined back to a person. Player-shared decks are a separate thing, and they live at /decks.

Add ?archetype=<key> to drill into one. That adds the archetype's own card performance, its matchup row, and cards_seen: every card recorded across its decks, which is a wider pool than the representative list — a card played in most builds but absent from that one deck would otherwise be invisible.

It is rate limited to a small number of requests per IP per day, because the snapshot is only regenerated about once a day — one call gets you everything, and a second call the same day almost always returns identical bytes. Read generated_at to tell whether it actually moved, and cache the rest. Over the limit you get a 429 with X-RateLimit-Limit.

Unknown scope or window values are rejected with a 400 rather than quietly falling back to the default — asking for ranked numbers and silently receiving the whole population is the one mistake that would not announce itself.

Card fields, in the order they appear in list responses: slug, set_code, card_number, name, pal_name, card_type, subtype, color, cost, power, strike, rarity, effect_text, keywords, work_keywords, is_lucky, image_url, thumbnail_url, status. The single-card endpoint adds japanese_name, durability, effect_text_ja, flavor_text, attack_name, orientation, art_url_full, revealed_at, source_url, usage_count.

For scripts, bots, and AI agents, the same API is described in machine-readable form:

• /openapi.json OpenAPI 3.1 description of every endpoint.
• /.well-known/api-catalog RFC 9727 catalog linking the spec, docs, and terms.
• /.well-known/agent-skills/index.json task-oriented skills (search cards, look up a card, browse decks).
• /llms.txt a plain-text site map for language models.

This page and the homepage also support Markdown content negotiation: send Accept: text/markdown to get a Markdown view instead of HTML.

The API surfaces the same data that powers the site itself. You're free to use it in personal projects, community tools, Discord bots, and open-source stuff. If you build something cool, tell us so we can link it. Please don't:

• Re-serve a copy of our card database as your own competing site.
• Scrape the API to train a model without asking us first.
• Attribute card art to yourself. All card art belongs to Bushiroad / Pocketpair.

See our terms of use for the full editorial stance.