KH Merch Graph

io.github.marleyhereicome-cpuPublished by GitHub account marleyhereicome-cpuv1.0.1Updated Nov 6, 2026

Identifies Kingdom Hearts merch from Japanese listing titles: SKU, condition terms, bootleg flags.

VerifiedStreamable HTTPWeb executableData & AnalyticsBusiness & Commerce

Overview

AI-generated overview

Remote MCP server that interprets Japanese secondhand Kingdom Hearts merch listings into SKU candidates, condition terms, bootleg flags and landed-cost…

What you can do
  • Resolve listing titles and descriptions into SKU candidates, condition terms, bootleg-caution flags and price-to-MSRP ratios, singly or in batches of up to five
  • Look up product details for a SKU or free-text query, including line, character, dimensions, rerelease history and source link
  • Browse the catalog by IP, character, budget and purpose
  • Estimate landed cost by proxy and shipping route for a price, weight or SKU and destination country
  • Report catalog errors or propose new SKUs with a supporting source
When to use it
Useful when you shop Japanese secondhand marketplaces for Kingdom Hearts merchandise and want an assistant to identify what a listing actually is, judge the price against MSRP, spot authenticity red flags, or estimate total cost shipped to your country. Only Kingdom Hearts is covered for now.
Requirements
A remote Streamable HTTP MCP endpoint; no API key or authentication is required. Any MCP-capable client can connect by URL, and a browser interface is also offered.
Before you install
Authenticity flags are cautions, not verdicts, and prices and landed-cost totals are planning estimates, not appraisals or quotes. The server does not scrape marketplaces; it reads only the listing text you send, and it says listing text and URLs are not logged, though tool name, matched SKU IDs, price, destination country and timestamp are logged, plus non-identifying keywords when nothing matches. report_correction submits data to the maintainers.

Installation

In SourceWeft

  1. Open KH Merch Graph in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Web executable via Streamable HTTP. Remote servers run from the web runtime once configured in a workspace.

Use with an agent

Paste this prompt into Claude Code, Codex, Cursor or another agent — or into SourceWeft chat. The agent reads this server's install guide, shows you its publisher, endpoint and sign-in, and adds it to your client once you agree. Keys never go into a command.

Read https://sourceweft.com/mcp/io.github.marleyhereicome-cpu%2Fkh-merch-graph/install.md and connect the MCP server it describes to my client. Before changing any config, show me its publisher, endpoint and sign-in method and wait for my OK. Never put an API key in a command or in a file I would commit; ask me where it should come from.

Read the install guide the agent follows

Other MCP clients

Add this to your client's mcpServers config.

{
  "mcpServers": {
    "kh-merch-graph": {
      "type": "http",
      "url": "https://mcp.merchgraph.app/mcp"
    }
  }
}

README

KH Merch Graph

An MCP (Model Context Protocol) server that turns a Japanese secondhand-listing title for Kingdom Hearts merchandise (Mercari, Yahoo! Auctions, Suruga-ya, Mandarake, …) into: the likely official SKU, what the condition wording actually means, bootleg-caution flags, a price-to-MSRP ratio, and a landed-cost estimate to your country. Point any MCP-capable AI agent at it and hand it listing text — it does the rest.

Try it in your browser (no AI agent needed): https://merchgraph.app

Why this is different

  1. Sourced catalog, not scraped or guessed. Every SKU in the catalog is tied to a primary source (an official product page, a maker's press release, etc.) and marked verified or not — the source URL is attached to every entry and returned with every match, so you can check it yourself. Nothing is inferred by an LLM inside the server.
  2. No confident guess, no candidate. When the match confidence is low, this server does not present a guess as an answer. Low-confidence matches are returned separately as weak_matches, never mixed into candidates — an empty candidates array means "probably not in the catalog," not "best effort."
  3. Bring Your Own Listing. This server never crawls or scrapes marketplaces itself, and never stores the listing text or URL you send it. You already have the listing open; you paste the text, it interprets it.

Supported IP

Kingdom Hearts only, for now. The catalog is curated one SKU at a time, so it grows slowly on purpose. If you ask about another franchise, the server won't fabricate an answer — it notes the request (see Data sourcing & privacy) to help decide what to add next. Use report_correction or open a GitHub issue if you want to nominate a franchise or a specific missing SKU.

Usage examples

These are things you'd say to your AI agent once it's connected to this server (see Connect).

1. "Is this listing worth it?"

"I found this on Mercari for ¥8,000: 一番くじ KINGDOM HEARTS -25th Anniversary- A賞 ソラ スタチュー 開封済み. What is it, and is that a fair price?"

The agent calls resolve_listing and gets back the matched SKU, the MSRP, the price ratio, the meaning of 「開封済み」("opened"), and a link to the primary source for that SKU.

2. "What Sora merch exists under ¥5,000?"

"Show me Kingdom Hearts Sora items under ¥5,000 that I could display."

The agent calls discover({ ip: "kingdom-hearts", character: "Sora", budget_jpy: 5000, purpose: "display" }) and gets a list of catalog SKUs matching those filters (this does not mean any of them are currently for sale anywhere — it's a catalog browse, not a marketplace search).

3. "What would it cost me, landed, in the US?"

"That listing is ¥8,000 plus it weighs about 400g. What's the total cost shipped to the US?"

The agent calls estimate_landed_cost({ price_jpy: 8000, weight_g: 400, dest_country: "US" }) and gets per-route estimates (proxy fee + shipping) with the cheapest option highlighted, plus when the rate table was last checked.

Connect

  • MCP endpoint (Streamable HTTP, JSON-RPC 2.0, no auth): https://mcp.merchgraph.app/mcp (the older https://kh-merch-graph.fandex.workers.dev/mcp keeps working)
  • Health check: .../health
  • For AI agents that discover servers by URL: .../llms.txt (plain-language entry point) and .../openapi.json (schema reference — the real transport is MCP's own tools/list/tools/call, this is a documentation aid for OpenAPI-only tooling)

Claude Desktop / Claude.ai

Add it as a custom connector using the MCP endpoint URL above. No API key or auth is required.

Any other MCP client

Point it at the /mcp endpoint with the Streamable HTTP transport. tools/list returns the live, authoritative schema for all five tools below.

Tools

ToolWhat it does
resolve_listingListing title/description → SKU candidates, condition terms, bootleg-caution flags, price ratio
resolve_listings_batchSame as resolve_listing for several listings in one call (up to 5 per call by default, set by MAX_BATCH_LISTINGS in wrangler.toml); results in input order
explain_productsku_id or free-text query → full detail on one SKU (line, character, dimensions, rerelease history, source)
discoverBrowse the catalog by IP/character/budget/purpose
estimate_landed_costListing price + weight/SKU + destination country → landed-cost estimate by route
report_correctionReport a catalog error, or propose a new SKU, with a primary source

Full input/output shapes: docs/SPEC.md.

Data sourcing & privacy

  • Every catalog row requires a source_url to a primary source; unverified rows are marked verified=false rather than silently trusted.
  • This server does not scrape, browse, or log in to any marketplace. It only reads the text an agent sends it, for the duration of a single request.
  • Usage logs on the public deployment contain only: tool name, matched SKU IDs, price, destination country, and timestamp. The listing title, description, and any URL you pass are never logged or stored.
  • When resolve_listing finds no candidate, or discover is asked about an IP not yet in the catalog, the server logs a small, non-identifying signal (a handful of recognized keywords, or the requested IP name only — never the raw listing text) so maintainers know what to add next.

Disclaimers

  • Authenticity is a caution flag, not a verdict. flags point out things worth asking the seller about (price far below MSRP, materials/marks associated with counterfeits, etc.). This server never states that an item is genuine or fake.
  • Prices are estimates for planning, not appraisals or quotes. MSRP ratios and landed-cost totals are meant to help you decide whether to look closer, not to be relied on for a purchase decision or a valuation.

Reporting a correction or a new SKU

Call report_correction with the field you want changed (or leave sku_id empty to propose a new item), your proposed value, and — importantly — a source_url backing it up. You can also open a GitHub issue.

Contributing / building this yourself

This repo is also a step-by-step teaching kit for building an MCP server like this one from scratch. See docs/GETTING_STARTED_JA.md (Japanese) and docs/SPEC.md for the full design. License: MIT (see LICENSE). Third-party names, logos and images are not covered; see NOTICE.

Pipeline: active

Source: README.md at commit c97a873

Tools

6

resolve_listing

Resolve listing

日本語の中古出品タイトル・説明文から、正規SKU候補・状態語・海賊版注意フラグ・定価比の目安を返す。

Parameterssrcurltitlerequiredplatformprice_jpydescription

resolve_listings_batch

Resolve listings (batch)

resolve_listing を最大5件まとめて1回で実行する。多数の出品を照会するときに使う。結果は入力と同じ順番の配列で、各要素は resolve_listing と同じ形式（失敗した件のみ { error }）。5件を超える場合は5件ずつに分けて複数回呼ぶこと。1件だけなら resolve_listing を使う。

Parameterslistingsrequired

explain_product

Explain product

sku_id または自然文クエリから、商品ライン・キャラ・寸法・再販履歴・出典URLなど商品の詳細を返す。

Parametersquerysku_id

discover

Discover

カタログから条件（IP・キャラ・予算・用途）に合う候補SKUを探す。実際に出品されているかは保証しない。

Parametersiprequiredlimittermspurposecharacterbudget_jpy

estimate_landed_cost

Estimate landed cost

出品価格・重量（またはsku_id）・仕向国から、代行×配送の経路別の総額目安と最安経路を返す。

Parametersproxysku_idweight_gprice_jpyrequireddest_countryrequired

report_correction

Report correction

カタログの誤り・不足を報告する。受付IDを返す。

Parametersnotefieldrequiredsku_idsource_urlproposed_valuerequired

Version history

1
  1. v1.0.1LatestOct 23, 2026

Score

100% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 daysRequired25 pts
  • Auth declaredRequired10 pts
  • Tools listedRequired15 pts
  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts