MCP server | Soldgraph Docs

Soldgraph runs a hosted Model Context Protocol server. An AI assistant connected to it can search sold and active listings, check card prices and see your remaining requests, with no code.

URL: https://api.soldgraph.com/mcp
Auth: your API key in the Authorization: Bearer sg_… header
Transport: Streamable HTTP, stateless

#Connect

Claude Code

BASH

claude mcp add --transport http soldgraph https://api.soldgraph.com/mcp \  --header "Authorization: Bearer $SOLDGRAPH_API_KEY"

Cursor (~/.cursor/mcp.json), VS Code (.vscode/mcp.json, under "servers") and other clients that take a JSON config:

JSON

{  "mcpServers": {    "soldgraph": {      "url": "https://api.soldgraph.com/mcp",      "headers": { "Authorization": "Bearer sg_your_key" }    }  }}

Keep the key in your client's config or an environment variable, never in a prompt or a shared repo.

#Tools

There is one tool for each search route. Each takes that route's parameters:

Tool	What it returns
`ebay_sold`, `poshmark_sold`, `mercari_sold`	Sold listings: what items actually sold for
`ebay_listings`, `poshmark_listings`, `mercari_listings`, `depop_listings`, `whatnot_listings`	Active listings and asking prices
`offerup_listings`, `offerup_item`	Local OfferUp listings near a ZIP code, and one listing's details
`tcgplayer_search`, `tcgplayer_sales`	TCGplayer products with market price, and a product's recent sales sample
`account_usage`	Your plan, requests used and remaining. Free
`api_status`	Each marketplace's live health. Free
`get_result`	Finishes a search that was still collecting. Free

All tools are read-only.

#How calls work

One call, one finished result. The server sends the search and waits for it, so the assistant gets rows instead of a job to poll.
Slow searches. If a search is still collecting after about 55 seconds, the tool returns a request_id. The assistant calls get_result with it, and that is not charged again.
Compact rows. Results include the fields most agents need: title, price, condition, dates and link. Pass fields: "full" for every field the API returns.
Same costs as the API. Each completed search costs 1 request, cache hits included. Failed searches, errors, get_result, account_usage and api_status are free. See Usage and billing.
Retries. A repeated identical call within five minutes reuses the same search, so a client retry is not charged twice.
Rate limits. Your plan's per-minute limit applies. A tool call uses one to four API calls of it.

#Example prompts

"What has a Nintendo Switch OLED sold for on eBay and Mercari this month? Give me the median and link three examples."
"Find used road bikes under $400 on OfferUp near 94103."
"Look up Charizard ex from Obsidian Flames on TCGplayer and show its recent sales."
"How many Soldgraph requests do I have left?"