Mtg Export

MTG Arena Collection Export

This skill captures the user's owned MTG Arena collection and drops it into their MTG workspace as collection/MTGA-export-<YYYY-MM-DD>.csv — the exact shape the mtg-std-build and mtg-std-upgrade skills read to prefer cards the user already owns. It runs in one shot, with no anchor cards and no manual data entry: it reads the collection directly out of the running game's memory and translates the Arena card ids to names using the local Scryfall database (mtg-db), so it never downloads or builds its own card data.

How it works

MTG Arena keeps the owned collection in memory as a map of arena_id → owned count. The bundled scanner (scripts/export_collection.py) attaches to the live MTGA process, finds that map anchor-free (it keeps the largest contiguous block of valid arena_id → count entries, validated against the card database), and writes the result as CSV. Card names/sets come from the arena_cards table in <workspace>/database/cards.sqlite (built by mtg-db); MTGA's own bundled card files, if present, are merged in best-effort so cards from a set newer than the database still resolve.

• Platforms: Windows (reads memory via pymem) and macOS (via Mach task_for_pid — needs sudo; works for the native Steam app and CrossOver/Wine). Linux is not supported.
• Dependencies: numpy (both platforms) and pymem (Windows only). See scripts/requirements.txt. These are the one exception to the repo's stdlib-only rule — reading another process's memory requires them.

Prerequisites (tell the user, once)

1. MTG Arena is installed and running.
2. In-game, open the Collection or Decks tab and scroll through the cards for ~30s so the whole collection is loaded into memory. (If the export comes back partial or empty, this is almost always why — scroll and re-run.)
3. The local Scryfall database exists with the Arena map. If the database is missing or was built before this skill existed, build/refresh it first via mtg-db (see below).

How to drive it (the one-shot)

Three steps — ready the card data, run the exporter, then push via mtg-sync. The export script does not push; the orchestrating skill invokes mtg-sync afterwards, as the normal user, so git uses the user's SSH key and config (see Sync below for why this matters):

1. Ensure the local Scryfall database exists and carries the Arena-id map (the arena_cards table). Do this via the mtg-db skill: check its status and, if the database is missing or was built before this skill existed, refresh it (a one-time ~78 MB download, ~30 s). The deck skills auto-build this database too, so on a machine that has already built a deck it's usually present — it just needs a refresh if it predates the Arena map.

2. Export with the bundled script:

   python "${CLAUDE_SKILL_DIR}/scripts/export_collection.py"
   

   It writes the CSV and prints its path to stdout. If the exporter exits with "No Arena card map in the Scryfall database", the database predates this skill: refresh it via mtg-db and re-run.

3. Push by invoking the mtg-sync skill (best-effort — see Sync below).

macOS: the scan needs root

On macOS, reading game memory requires root. Plain sudo needs a TTY, which a non-interactive agent shell doesn't have (sudo: a terminal is required to read the password). Use the bundled askpass helper so sudo -A pops a native, secure macOS password dialog instead — the user types the password into a GUI prompt, with no terminal and nothing in the transcript:

SUDO_ASKPASS="${CLAUDE_SKILL_DIR}/scripts/macos-askpass.sh" \
  sudo -A python3 "${CLAUDE_SKILL_DIR}/scripts/export_collection.py"

(If you'd rather type the password in a terminal, the plain sudo python3 …export_collection.py form still works when run interactively, e.g. via the ! prefix in Claude Code.)

When run under sudo, the script chowns the written CSV back to the invoking user (via SUDO_UID/SUDO_GID), so exports never land root-owned and the later mtg-sync push can stage them cleanly.

(When installed via the plugin marketplace, ${CLAUDE_SKILL_DIR} resolves to this skill's directory. The script locates the shared mtg_scryfall library itself, so a manually-copied layout works too as long as the skills sit side by side.)

Flags

--db PATH            Use an explicit cards.sqlite (default: the resolved workspace database).
--out-dir DIR        Write the CSV somewhere other than <workspace>/collection/.
--process NAME       Override the MTGA process name (default: MTGA.exe / MTGA / mtga).

The script prints progress to stderr and the single written CSV path to stdout. Exit codes: 0 success · 2 setup/dependency problem (numpy missing, no Arena map in the DB) · 3 could not locate the collection (game not running, or cards not loaded — scroll and retry).

Output

• <workspace>/collection/MTGA-export-<YYYY-MM-DD>.csv — Moxfield-format CSV with the columns Count, Name, Edition, Condition, Language, Foil, Tag (Condition Near Mint, Language English). One row per (name, set), counts summed. This is exactly what the Arena deckbuilding skills look for under collection/.

The workspace is resolved the usual way: $MTG_HOME → nearest .mtg/ → ./.mtg/ (see the mtg-db and mtg-sync skills). Point MTG_HOME at a synced mtg-data repo and the export lands there directly.

Sync (run after every export)

Once the CSV is written, push it by invoking the mtg-sync skill — committing and pushing the collection so it's available on the user's other machines and to the deckbuilding skills. The export script never pushes itself. On macOS the scan runs under sudo, and a git push from inside that process would run as root — which has neither the user's SSH key nor their git identity, so the push fails with Permission denied (publickey). Running mtg-sync separately, as the normal user, sidesteps that entirely.

This push is best-effort, exactly like the deck skills' push step: if the workspace isn't a git repo, git isn't installed, or the network is down, mtg-sync reports it in one line and the CSV still sits in the local workspace to be pushed later — the export itself still succeeded, and the deckbuilding skills read the CSV from the local workspace regardless.

Troubleshooting

• "Collection not found" / partial export → MTGA wasn't fully loaded. Open the Collection/Decks tab, scroll through the whole collection, then re-run.
• macOS "Cannot access game memory" → re-run under sudo. From a non-interactive shell use the askpass helper (SUDO_ASKPASS=…/macos-askpass.sh sudo -A python3 …) so the password prompt is a GUI dialog rather than a TTY; see macOS: the scan needs root above.
• sudo: a terminal is required to read the password → you ran plain sudo from a shell with no TTY. Use the sudo -A + SUDO_ASKPASS form above.
• Windows permission error → run the terminal as Administrator.
• "No Arena card map in the Scryfall database" → the database predates this skill (or is missing). Run mtg-db with --refresh to rebuild it with the arena_cards table.
• macOS process not found → ensure MTGA is open (Steam or CrossOver); the scanner looks for MTGA, MTGA.exe, and mtga. Override with --process if needed.
• Unsupported platform → only Windows and macOS can be scanned; Linux is not supported.