swu-mcp

Star Wars: Unlimited card data, scores, synergy and rules, ready for your AI assistant to look up.

What it is

swu-mcp is a small program that an AI assistant such as Claude can talk to. It answers questions about Star Wars: Unlimited (SWU) cards: what a card says, how strong a rubric rates it, which cards work well together, what it costs, and what the official rules say. It speaks the Model Context Protocol (MCP), an open standard that lets assistants call outside tools.

Every number comes from fixed code run over saved inputs. No AI model runs when you ask a question, and none picks a score. The same data always gives the same answer. All 48 tools only read data; none changes anything.

Install for Claude Code

You need Node.js 22 or newer. Run this once:

claude mcp add swu -- npx -y github:LampaGJ/swu-mcp

The first start builds the server from GitHub and downloads the card data (about 2 MB). Both are kept for later runs.

Install for Claude Desktop

Open the Claude Desktop settings, choose Developer, then Edit Config. Add this entry to the mcpServers section and restart the app:

{
  "mcpServers": {
    "swu": {
      "command": "npx",
      "args": ["-y", "github:LampaGJ/swu-mcp"]
    }
  }
}

The 48 tools

Cards and search

get_card, search_cards, list_facet, cards_by_facet, get_cards_batch, resolve_card_name, get_printings

Value

score_card, rank_cards, compare_cards, explain_rubric

Text features

card_features, find_by_effect

Synergy and combos

find_pairings, find_trios, explain_synergy, aspect_cover

Fungibility and substitutes

fungible_substitutes

Market and play data

card_price, play_stats, value_vs_price

Rules

rules_core, rules_search, get_rule, card_rulings, keyword_rules, rules_version, search_rulings

Similarity

similar_cards, kin_cards, cluster

Sets and legality

list_sets, format_legality

Decks, collections and trades

deck_audit, deck_legality_check, suggest_leader_base, playable_cards, fungible_in_deck, budget_alternatives, price_list, trade_balance, owned_vs_needed, play_value_vs_price, score_vs_play

These take a list of cards you pass in. The server stores nothing and lists any name it cannot match.

Meta defense

meta_archetypes, meta_threats, meta_answers

The current tournament meta, grouped into archetypes from recent Premier tournament decklists, and the cards your aspects can play that answer the most of it. Win rates come from tournament results (swumetastats), with the number of matches behind each.

Provenance

data_version

Resources and prompts

Resources the assistant can read: swu://rubric, swu://glossary, swu://usage-guide (which tool answers which question), swu://rules/{section} and rules://core-primer.
Prompts: rules_question and build_deck_for_leader.

The README gives a one-line description of each tool.

Data version

This release uses card data built on 2026-10-16. The server checks the download against a fixed checksum before it uses it. See the data-2026-10-16 release. To use your own data folder instead, set SWU_MCP_DATA.

Honest limits

Pairings, trios and fungible substitutes are candidates, not facts. Sampled reviews found card-text features wrong on about 1 card in 5 and about 1 in 4 pairings wrong on a fixed 200-pair benchmark, so confirm a pairing against both cards’ text. The pairing, trio, explain and substitute tools return each card’s clause text with every candidate, marked status: candidate, so the assistant can check the link. Pairings and trios rank mechanical links first; pairs that only share a trait (obvious tribal links) are listed apart. Copied data (card fields, rulings, prices, play stats, rules sections) passed a full audit against the originals with no failures.
The play data is thin. It is online Karabast and Petranaki games in three Premier windows, not tournaments. Homeworlds has under one week of play data.
Prices mostly follow rarity. They are US TCGplayer prices from one import, so cheap commons dominate any value-per-dollar ranking.
The total score is comparable only within a card type. The playCalibrated score is rank-meaningful only, and it is fitted to the same play data it is compared with.
The check for a newer Comprehensive Rules version is manual. Nothing schedules it.