Skip to main content
Glama

PokémonGPT connects Claude, ChatGPT, Cursor, VS Code or any MCP client to a database of about 21,000 English Pokémon cards and 80,000 TCGplayer listings, so card details and prices are looked up instead of guessed. Ask things like:

  • "What's the most valuable card in Evolving Skies?"

  • "How much is my Base Set Charizard 4/102?"

  • "Cheap Fire ex cards under $5 for my deck"

  • "Prismatic Evolutions Elite Trainer Box price"

  • "喷火龙最贵的卡是哪张?"

  • "日版喷火龙多少钱?"

From the GPT Store

PokémonGPT began as a custom GPT in the GPT Store, where it was rated 4.3★ by 70+ users across 5,000+ conversations. Custom GPTs retire on December 11, 2026, so PokémonGPT continues under the same name and logo as:

  • An open MCP server that works in any assistant supporting MCP, not just ChatGPT

  • A portable skill with the original persona and how to search and present prices well

  • A ChatGPT plugin that bundles the skill and the MCP server for ChatGPT users

Related MCP server: tcgplayer-api-mcp

Connect

The hosted server is free to use and needs no sign-in:

https://pokemongpt-455505.uc.r.appspot.com/mcp

It is read-only Streamable HTTP, run on a best-effort basis: please be gentle with it, or host your own.

Client

Setup

Claude (web / desktop)

Settings → Connectors → Add custom connector → paste the URL

Claude Code

claude mcp add --transport http pokemongpt https://pokemongpt-455505.uc.r.appspot.com/mcp

ChatGPT

Install the PokémonGPT plugin, or add the URL as a custom connector where your plan offers developer mode

Cursor / VS Code

See the config below

Other clients

Any client that supports remote Streamable HTTP servers

Cursor, ~/.cursor/mcp.json:

{ "mcpServers": { "pokemongpt": { "url": "https://pokemongpt-455505.uc.r.appspot.com/mcp" } } }

VS Code, .vscode/mcp.json:

{ "servers": { "pokemongpt": { "type": "http", "url": "https://pokemongpt-455505.uc.r.appspot.com/mcp" } } }

The MCP server provides the data. The skills/pokemongpt skill teaches the assistant to use it well: translate names to English before searching, pick the right tool, report each printing's price separately, check that every price belongs to the card asked about, and add the right disclaimers. It also carries the original PokémonGPT persona for general Pokémon questions: games, Pokédex, competitive play and more.

  • Claude Code: copy skills/pokemongpt to ~/.claude/skills/

  • Claude (web / desktop): zip the folder (cd skills && zip -r pokemongpt.zip pokemongpt) and upload it in Claude's skills settings

  • ChatGPT: included in the PokémonGPT plugin

Without an MCP connection, the skill falls back to scripts/tcg_api.py, a standard-library client for the same API.

Tools

Tool

Use it for

search_cards

English cards by name, set, number, rarity, type, stage, tag (ex, VMAX, ...), attack or ability text, illustrator, regulation mark, HP or price. Returns card details and the price of every printing

search_prices

Raw TCGplayer listings: sets too new to have card details yet, Japanese cards (language="jp") and sealed product (singles_only=false)

list_sets

English sets, newest first, and which are only covered by search_prices

list_filter_values

The exact values filters accept: rarities, stages, tags, ...

Every tool is read-only. The same searches are also a plain HTTP API: see /docs on the server.

Data

Source

Coverage

Card details

PTCG-database (MIT), built from pkmncards.com

About 21,000 English cards across almost every set

Prices

TCGplayer, via tcgcsv.com

About 47,000 English and 33,000 Japanese listings, singles and sealed

  • Prices are TCGplayer market prices in USD for ungraded (raw) cards, not PSA/BGS/CGC graded values. Every response includes prices_as_of; prices are refreshed periodically, not live.

  • About 96% of cards are matched to a TCGplayer listing by set, number and card name. Most of the rest are trainer kits and TCG Classic decks, which TCGplayer lists differently.

  • A set released in the last few weeks may only be in search_prices until the card database catches up.

Self-hosting

You need Python 3.12 and a MongoDB database. A free MongoDB Atlas cluster is plenty: the data is about 60 MB.

git clone --recurse-submodules https://github.com/danielqingz/pokemongpt-mcp.git
cd pokemongpt-mcp
python3.12 -m venv .venv && .venv/bin/pip install -r requirements.txt
cp .env.example .env                                 # then set MONGO_URL

.venv/bin/python scripts/fetch_prices.py --lang en   # TCGplayer prices, 10+ minutes
.venv/bin/python scripts/fetch_prices.py --lang jp
.venv/bin/python scripts/load_data.py                # cards + prices into MongoDB, ~2 minutes

.venv/bin/uvicorn server.main:app --port 8000        # MCP at http://localhost:8000/mcp, API docs at /docs

To update later, run git submodule update --remote data for new cards, then scripts/load_data.py --refresh-prices.

Deploying. app.yaml is ready for Google App Engine (gcloud app deploy; .env is uploaded with the app). It keeps one instance warm, because a cold start takes about 25 seconds and MCP clients can time out waiting, and caps the server at 2 instances, because the endpoint is public and needs no sign-in. Elsewhere, run the same uvicorn command. If your database only accepts listed IP addresses, allow the host's.

macOS certificates. The python.org installer ships without root certificates. If downloads or Atlas connections fail with CERTIFICATE_VERIFY_FAILED, run export SSL_CERT_FILE=$(.venv/bin/python -m certifi) first.

ChatGPT plugin. python scripts/build_plugin.py packages plugin/ and skills/ into dist/pokemongpt-<version>.zip after checking the listing limits from OpenAI's submission docs. Point plugin/mcp.json and the URLs in plugin/plugin.json at your own server first. plugin/REVIEW.md lists the remaining submission steps.

Layout

server/main.py         FastAPI app: HTTP API (/tcg/...) and the MCP endpoint (/mcp)
server/mcp_server.py   MCP tools
server/service.py      query logic shared by both (typed filters only, no raw database queries)
server/models.py       parameter and response models
scripts/               fetch_prices.py (TCGplayer via tcgcsv), load_data.py (MongoDB loader), build_plugin.py
skills/pokemongpt/     the skill: SKILL.md, reference.md, scripts/tcg_api.py
plugin/                ChatGPT plugin manifest, MCP config, logo, submission checklist
data/                  card database (git submodule)

Privacy

The hosted server has no accounts and stores no searches. Like any web server, its host (Google Cloud) keeps standard request logs (IP address, time, requested URL) for a limited period. Your AI assistant's own privacy policy covers your conversations. Full policy: pokemongpt-455505.uc.r.appspot.com/privacy.

Disclaimer

PokémonGPT is an unofficial fan project. It is not affiliated with, endorsed by or sponsored by Nintendo, The Pokémon Company, Creatures, GAME FREAK or TCGplayer. Pokémon and all related names are trademarks of their respective owners. Card text and images belong to their copyright holders. Prices are informational only and change constantly: check several sources before buying or selling.

License

MIT for the code in this repository. Card data and images keep their own licenses and terms.


中文说明

PokémonGPT 原本是 GPT Store 上的宝可梦助手,获得了 70 多位用户 4.3 星的评分,累计 5000 多次对话。自定义 GPT 将于 2026 年 12 月 11 日下线,PokémonGPT 沿用原来的名字和 Logo,以开源 MCP 服务器、技能(skill)和 ChatGPT 插件的形式继续提供服务。

  • 直接使用: 把 https://pokemongpt-455505.uc.r.appspot.com/mcp 添加到 Claude、ChatGPT、Cursor 等支持 MCP 的工具里(方法见上方 Connect),再安装 skills/pokemongpt 技能,效果更好。ChatGPT 用户也可以直接安装 PokémonGPT 插件。

  • 可以用中文提问,比如"喷火龙最贵的卡是哪张?""日版喷火龙多少钱?"。助手会先把卡名翻译成英文再查询。

  • 价格说明: 来自 TCGplayer 的美元市场价,指未评级的散卡(不是 PSA 等评级卡),每次结果都会注明价格日期。

  • 自己部署: 见上方 Self-hosting。

  • 本项目为非官方粉丝作品,与任天堂、宝可梦公司及 TCGplayer 无关。

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • F
      license
      Not graded
      quality
      D
      maintenance
      Provides access to TCGplayer trading card data, including search, product details, pricing, and market information, enabling natural language queries for card analysis.
      1
      -
    • F
      license
      Not graded
      quality
      B
      maintenance
      Enables AI assistants to look up trading-card market values across Pokémon, Magic, sports and other TCG catalogs, including raw prices by condition, graded ladders, price history, trending movers and set checklists. It also calculates grading ROI — gem premiums, net profit after fees, expected value and break-even gem rates — so collectors can decide whether a card is worth submitting.
      -