Pokémon Champions MCP Server

An MCP server (TypeScript, stdio) that gives an AI client accurate, current information for competitive Pokémon Champions (the VGC-style battle game). It exposes tools for damage calculation, type effectiveness, Pokémon lookup, and regulation legality (e.g. which Pokémon are allowed in Regulation M-B).

Datasets

The server reads three local datasets at startup and never touches the network on a normal tool call — only the on-demand scraper scripts do.

1. Damage engine — the damage formula and its mechanics (abilities like Contrary, weather, items, crits, multi-hit, spread reduction) come from @smogon/calc, whose engine descends from the same Honko/Zarel codebase the trusted community calculators use. We wrap it; we never reimplement the formula. (Generation 9.)

2. Champions engine dex — per-Pokémon facts (base stats, typing, abilities, weight, and the full learnable movelist), in data/champions-dex.json, scraped from Serebii's Champions Pokédex (207 Pokémon, 282 forms incl. Megas, ~62 moves each). Why not just use @smogon/calc for this? Because @smogon/calc ships mainline data, which is wrong or missing for Pokémon Champions' game-original Mega Evolutions — e.g. Champions' Mega Staraptor has Contrary, but @smogon/calc says Intimidate; and forms like Mega Eelektross don't exist there at all. get_pokemon uses this dex as the source of truth, and calculate_damage overrides @smogon/calc's species data with it (auto-filling stats/typing/ability) so even Champions-original Megas calc correctly. This is a deliberate deviation from the original "never scrape engine data" rule, made because the package doesn't match the game.

3. Legality data — which Pokémon are legal in a specific Champions regulation, in regulations/. Small, fast-changing, Champions-specific, in no package.

Related MCP server: Pokémon MCP Server

Build

npm install
npm run build      # tsc -> dist/

Requires Node 18+ (uses built-in fetch in the scraper).

Optional sanity check — runs all five tools through an in-memory MCP client:

npm test

Register in Claude Code

Add this to your Claude Code MCP config (use an absolute WSL/Linux path — Claude Code here runs via AWS Bedrock inside WSL2):

{
  "mcpServers": {
    "pokemon-champions": {
      "command": "node",
      "args": ["/home/emielkoridon/git_repos/poke-mcp-tool/dist/index.js"]
    }
  }
}

Then restart Claude Code. (Run npm run build first so dist/index.js exists.)

Tools

Refresh the data

Both scrapers are standalone and run manually — the server never invokes them, it only reads their JSON output.

Regulation legality (the full legal roster + bans for a regulation):

npm run scrape -- m-b

This reads two Serebii pages: the regulation page (dates + what's newly useable) and the Champions Pokédex index (the full legal roster — all HOME-transferable Pokémon). It writes regulations/m-b.json with legalPokemon (the full roster, 207 for M-B), bannedPokemon (format-excluded Mega forms — for M-B, Mega Garchomp Z & Mega Lucario Z), and meta (dates, source URLs, newlyUseable, scrapedAt). Parameterize by id for future regulations (npm run scrape -- m-c). Per the official rules there are no Restricted Pokémon, so everything in the dex is legal except the listed Mega exclusions.

Champions engine dex (per-Pokémon stats/types/abilities — only needed when the game adds Pokémon/forms):

npm run scrape:dex                 # full roster (~207 pages, throttled — takes a minute)
npm run scrape:dex -- staraptor    # one or more slugs, for testing (writes a .sample.json)

Both scrapers are defensive: if Serebii's markup changes or a page is only partially published, they print a loud warning and write what they found rather than crashing. The brittle, page-specific selectors are isolated in parseRegulationPage() / parsePokemonPage() so they're easy to fix.

Known data limitations (surfaced honestly, not faked)

These come from the real source data and are reported in the tool output, not hidden:

• The format publishes no per-move bans, so move legality = whether the Pokémon can learn the move in Champions (its scraped learnset). check_legality and get_pokemon state this; a move marked "legal" means learnable, not separately whitelisted.

• Mega legality follows the base species. A Mega isn't a separate roster entry, so check_legality("Mega Staraptor") resolves via base "Staraptor" (and the result note says so). Specific Mega forms in bannedPokemon (Mega Garchomp Z, Mega Lucario Z) are excluded even though their base species is legal.

• Champions-original ability effects aren't modelled by the damage engine. @smogon/calc computes base damage from the correct (overridden) stats/types, but it can't simulate the special effect of an ability it doesn't know (e.g. Mega Eelektross's "Eelevate"). Stats, typing, and the named ability are Champions-accurate; the ability's mechanical effect on the calc may be ignored for game-original abilities.

• Regional-form typing can be noisy. A few Pokémon that share a dex page with a regional form (e.g. base Raichu vs Alolan Raichu) may show the union of both forms' type links in the base entry. Mega forms — the competitively relevant part — are parsed cleanly per form.

Project layout

src/index.ts                    MCP server: declares the 5 tools + stdio transport
src/calc.ts                     wraps @smogon/calc (damage formula + type effectiveness)
src/dex.ts                      get_pokemon lookups over data/champions-dex.json (fallback @smogon/calc)
src/regulations.ts              loads/validates regulations/*.json, legality checks
src/names.ts                    shared Pokémon-name canonicalization (Mega word order, base species)
scripts/scrape-champions-dex.ts standalone scraper -> data/champions-dex.json (engine dex)
scripts/scrape-regulation.ts    standalone scraper -> regulations/<id>.json (legality)
data/champions-dex.json         Champions engine dex (generated; committed)
regulations/m-b.json            legality data (generated; committed)
test/smoke.ts                   in-memory MCP client exercising all 5 tools