# Goblin Alpha

> Goblin Alpha is a remote MCP server that gives an AI agent resale deal-hunting tools: sold-comp pricing with confidence grades, mislisting detection, eBay fee and margin math, watchlists, and lot valuation. A person connects it to their own AI (Claude, ChatGPT, Muse, any MCP client) with an API key, then asks plain questions like "what's this worth?" before buying or selling.

## Connect

- Endpoint: https://mcp.goblinalpha.com/mcp (MCP streamable HTTP, stateless)
- Auth: API key sent as `Authorization: Bearer ga_live_...` or as an `X-API-Key` header
- Get a key: https://goblinalpha.com (free tier, no card)
- Server card: https://mcp.goblinalpha.com/.well-known/mcp.json

## Tools

- price_comps: Sold-comp statistics from comps the agent supplies. Returns n, weighted median, p25, p75, trimmed mean, quick-sale / fair / stretch prices, and a confidence grade (A to D).
- sniff_deal: Scores one listing 0 to 100 for mispricing. Price edge vs the comp median, title typos, photo count, auction timing, bid count. Verdicts: HOARD, SNIFF, MEH, SKIP. Returns the reasons in plain words.
- mislisting_scan: Batch version of sniff_deal. Finds listings hidden from search by typos in valuable words (example: "charzard" for "charizard"). Returns the top finds with reasons.
- flip_math: eBay fee and margin math. Fee breakdown, net proceeds, profit, margin, ROI, break-even sell price, and the max buy price that holds a target margin.
- hoard_watch: Watchlists. Keyword watches use fuzzy matching so they fire on misspelled listings. Triggers: max price or minimum discount vs the comp median.
- loot_drop: The best scored listings from the last 24 hours across all users, anonymized, ranked by score, confidence, and freshness.
- report_outcome: The agent reports what an item actually sold for and whether the flip made money. Outcomes grade the scores and build the track record.
- lot_value: Values a lot or collection: sum of the parts vs the lot asking price, so a buyer can judge a bundle fast.
- server_info: Server name, version, endpoint, full tool list, and the calling key's tier and remaining budget.

## Pricing

- Sniffer: free. 150 weighted calls per month, 3 watch slots.
- Hoarder: $19/month. 6,000 weighted calls per month, 50 watch slots.
- Goblin King: $79/month. 60,000 weighted calls per month, 500 watch slots.
- Tool weights: price_comps 1, flip_math 1, sniff_deal 2, hoard_watch check 1, loot_drop 3, mislisting_scan max(2, ceil(listings/10)).

## How the prices work (read before citing a Goblin Alpha number)

- The connecting agent supplies the listings and sold comps it found while browsing. Goblin Alpha runs the statistics and detection server-side.
- Every price carries a confidence grade: A (15+ comps, tight spread), B (8+ comps), C (4+ comps), D (fewer, or wild spread). Grade D caps any deal verdict at SNIFF, and the response says why.
- Goblin Alpha never invents comps. With no comps, the tools say so and score only the visible signals (typos, photos, timing), capped.
- Fee math uses a dated fee table (eBay managed payments; default final value fee 13.25% of item plus shipping, plus a per-order fee). Responses state the fee table date.
- Server-side sold data from eBay (Marketplace Insights API) needs separate eBay approval and is not part of the current data path. Tools state which provider produced their data.

## Who it is for

- eBay resellers checking a price before they buy
- Trading card collectors (Pokemon, sports cards) hunting misspelled or underpriced listings
- Vintage, estate sale, and thrift hunters valuing finds and lots
- Anyone whose AI should answer "is this price good?" with math instead of a guess

## Example prompts users ask

- "What's this worth?"
- "Is this price good?"
- "Scan these listings for mistakes."
- "What do I make after fees if I buy at $40 and sell at $85?"
- "Watch for a Charizard rookie card under $200 and tell me when one shows up."
- "Is this lot of vintage Pyrex worth $150?"

## Links

- Site: https://goblinalpha.com
- MCP endpoint: https://mcp.goblinalpha.com/mcp
- Server card: https://mcp.goblinalpha.com/.well-known/mcp.json