Myfigurecollection API MCP server

myfigurecollection-api

An unofficial MyFigureCollection.net API: a Python library plus a local MCP server, so agents can read figure data, collections, lists and clubs.

MFC has no working public API. Its own API club (#349) has been promising "API v4 coming soon" for eight years. This package reads the public HTML instead, and turns it into typed models.

New here? The short version

MyFigureCollection is the largest catalog of anime figures and merch on the internet, and for years there has been no good way for a program to read it. This project fixes that on your own computer. You install it once, and an AI assistant like Claude can then look things up on MFC for you: "how much is this figure at AmiAmi?" or "list every Chiikawa item releasing in October."

sequenceDiagram
    participant You
    participant Claude
    participant api as mfc-api, on your computer
    participant MFC as myfigurecollection.net
    You->>Claude: How much is the Alice Carroll figure?
    Claude->>api: get_partner_listings(287)
    api->>MFC: fetch the item's Buy window
    MFC-->>api: HTML
    api-->>Claude: AmiAmi · Available · ¥11,980
    Claude-->>You: AmiAmi has it in stock for ¥11,980

What is MCP? The Model Context Protocol is a standard plug for giving AI assistants new abilities. An MCP server is a small program on your machine that an assistant is allowed to call. This one gives your assistant 12 MFC abilities, from item search to barcode lookup. Adding it takes one snippet in a config file, shown in Use it as an MCP server.

Why not use MFC's official API? There isn't one. The site's own API club has been waiting for it since 2018.

Why isn't this a website I can visit? Cloudflare guards MFC and decides who gets in partly by IP reputation. It trusts connections from home computers and distrusts servers, so a hosted version would get blocked within days while a copy on your own machine keeps working. That constraint shaped the whole design.

Do I need to know Python? Two terminal commands to install (below). After that your assistant does the driving.

Why this exists (and why the obvious approach fails)

MFC is behind Cloudflare, which blocks on TLS fingerprint — not user-agent. requests, httpx and aiohttp all get a 403 "Just a moment..." challenge no matter what headers you send. That is why the older tenji scraper stopped working; its parsers never even received HTML.

This package uses curlcffi with impersonate="chrome", which reproduces a real Chrome handshake and gets a 200. That choice is load-bearing.** Swapping in a normal HTTP client will break everything.

It is also why this ships as a local tool rather than a hosted service: Cloudflare weights IP reputation, so a scraper on datacenter IPs gets challenged and burns the shared address for everyone. Run it on your own machine.

flowchart LR
    A["requests / httpx / aiohttp<br/>(any headers you like)"] -->|"403 Just a moment..."| CF{Cloudflare}
    B["Your browser"] -->|200| CF
    C["mfc-api via curl_cffi<br/>(Chrome TLS handshake)"] -->|200| CF
    CF --> MFC[("myfigurecollection.net")]

Install

This needs Python 3.10+. The package is on PyPI:

pip install myfigurecollection-api

On a Mac, python3 is often Xcode's 3.9, which will not run this code, so be explicit about the interpreter (e.g. pip3.12). Working from a clone instead, use pip3.12 install -e ..

Then check it landed:

mfc-api item 287

If mfc-api isn't found, your shell's python3 bin directory isn't on PATH. Find the script with python3.12 -c "import sysconfig; print(sysconfig.get_path('scripts'))" and use that absolute path.

Use it as an MCP server

Shortened. The full README is on GitHub.