MCP MCP server

eBay MCP server. Sold prices for your AI.

Connect Claude, Cursor, Gemini or VS Code to api.soldgraph.com/mcp and ask what anything sold for. Real eBay, Poshmark and Mercari sales with links, plus live listings on six more marketplaces.

100 free requests · No credit card required

Terminal
claude mcp add --transport http soldgraph https://api.soldgraph.com/mcp \  --header "Authorization: Bearer $SOLDGRAPH_KEY"
See it work

One URL and a key. Then just ask.

The server waits for each search and returns compact rows with a price summary, so the assistant answers in one step.

Ask in plain EnglishExample data

What have used Switch OLEDs sold for on eBay lately?

ebay_soldq: "nintendo switch oled", condition: ["used"]

200 recent sales on one page. Median $180; bundles and special editions sold higher.

  • Switch OLED, white, dock and boxSold Sep 29$185
  • Switch OLED neon, console onlySold Sep 28$164
  • Switch OLED with case and 2 gamesSold Sep 27$229
Terminal
claude mcp add --transport http soldgraph https://api.soldgraph.com/mcp \  --header "Authorization: Bearer $SOLDGRAPH_KEY"
One URL and your API key. Every tool is read-only.Claude Code setup
Set up your assistant

Pick your client. Each guide takes one step.

Claude Code

One terminal command. Price a list, research comps or build a script that calls the API.

Cursor

Add one block to mcp.json. Ask the agent for sold prices while you build your app.

Claude Desktop

Add the server to Claude's config and ask in chat what something is worth.

Gemini CLI

Install the Soldgraph extension with one command. Gemini asks for your key and stores it securely.

VS Code and Copilot

Add .vscode/mcp.json. VS Code asks for your key once and stores it securely.

Tools

19 tools, all read-only.

One tool per search route. A completed search costs one request, the same as the API. Failed searches and the last three tools are free.

ToolWhat it returns
ebay_soldeBay sold listings
ebay_listingseBay active listings
poshmark_soldPoshmark sold listings
poshmark_listingsPoshmark active listings
mercari_soldMercari sold listings
mercari_listingsMercari active listings
depop_listingsDepop active listings
whatnot_listingsWhatnot active listings
tcgplayer_searchTCGplayer product search
tcgplayer_salesTCGplayer recent sales
offerup_listingsOfferUp local listings
offerup_itemOfferUp listing details
facebook_listingsFacebook local listings
facebook_itemFacebook listing details
vinted_listingsVinted active listings
vinted_itemVinted listing details
account_usageYour plan, requests used and remaining
api_statusEach marketplace's live health
get_resultFinishes a search that was still collecting

Full tool list, slow-search handling and rate limits are in the MCP guide.

Building with ChatGPT? Paste this.

One prompt gives your assistant the endpoint, the sign-in header and how to wait for a result, so its first script works.

Prompt for ChatGPT, Claude or Cursor
I'm building with the Soldgraph API. It returns sold and for-sale listings from eBay and other resale marketplaces.

Read the docs first: https://soldgraph.com/llms.txt
OpenAPI spec: https://soldgraph.com/openapi.json

Rules for the code you write:
- Base URL: https://api.soldgraph.com/v1
- Auth: send the header "Authorization: Bearer <key>". Read the key from the SOLDGRAPH_KEY environment variable. Never put it in browser code.
- eBay sold prices: GET /ebay/sold?q=<search words>. Add count=200 for up to 200 rows.
- A search often answers 202 with status "pending" and a poll_url. Request https://api.soldgraph.com plus poll_url with ?wait=20 and the same header, until status is "complete" or "failed".
- Read result.summary for the median, low and high price. The rows are in result.data, each with title, displayed_price, sold_date and link.
- Each completed search uses one request. A free account starts with 100 requests that do not refill, so save results while testing instead of repeating a search.

Start with a short script that prints the median sold price for "nintendo switch oled". Then ask me what I want to build.
Paste it in, add your key as SOLDGRAPH_KEY, and it writes working code.Open in ChatGPT Open in Claude
FAQ

Common questions

Something else? Email hello@soldgraph.com and a person will reply.

MCP (Model Context Protocol) is the standard way AI assistants call outside tools. Connect an assistant to an MCP server once, and it can use that server's tools whenever a question needs them.

Connected to Soldgraph, an assistant can look up what an item sold for on eBay and answer with real sales and links.

Not for new developers. eBay retired findCompletedItems in February 2025, its Browse API only returns active listings, and its sold-data API, Marketplace Insights, isn't open to new users. An eBay MCP server built on those APIs can list items for sale but can't tell you what they sold for.

Soldgraph's server returns eBay sold listings in 8 countries, plus sold listings on Poshmark and Mercari.

Claude Code, Cursor, Gemini CLI, VS Code with GitHub Copilot, Claude Desktop and the Claude API, plus any MCP client that supports Streamable HTTP with a custom header.

Agents that read skills, like Codex and OpenClaw, can install the Soldgraph skill with npx skills add soldgraph/soldgraph-agent.

ChatGPT connectors can't send an API key yet. To use Soldgraph with ChatGPT today, have it write code from the ready-made prompt in the docs.

The same as the API. Each completed search costs one request. Failed searches and the usage, status and get-result tools are free.

Start with 100 free requests, no credit card required. Plans start at $9 a month for 2,000 requests.

No. Every tool is read-only and searches public listings. It never asks for your eBay login.

They are the prices the marketplace shows. When an eBay sale closed on an accepted Best Offer, eBay does not publish the amount, so that row is flagged and the summary also gives the median without those sales.

Ask your AI what it sold for in about two minutes.

Create a free account, make a key, and send your first request. Start with 100 free requests, no credit card required.