sailorpepe/litvm-tcg-oracle-mcp

Plug any AI agent into 455K+ real trading card prices across 25+ games — every price backed by on-chain Merkle proofs, not blind trust.
 
 
 
 
🔌 Connect over MCP — one URL, no install
No install, no account, no API key. **16 tools** over streamable HTTP — all free.
The full oracle through the LitVM lens: prices, on-chain Merkle proofs (raw AND
graded slabs), calibrated forecasts with a public accuracy scorecard, card-collateral
loan-terms previews, sports boards, the slab census, and the 4,444-soul fantasy
league whose weekly lineups are committed to the PredictionRegistry on LiteForge
before games score. Paid endpoints (full loan quotes, per-player sports forecasts)
live on the flagship oracle at https://mcp.the-undesirables.com via x402.
**Claude Desktop / Perplexity** — add it as a custom remote connector (Perplexity:
Settings → Connectors → + Custom Connector → Remote).
**Cursor / Windsurf / VS Code** — clients that take a URL in config:
Tools include search_tcg_products, card_forecast, simulate_price,
recommend_workflow and oracle_scorecard — call tools/list for the live set.
Everything is free: pay-per-call has been paused since 2026-10-02, and AI grading is retired.
Search is set-aware — search_tcg_products("Base Set Charizard") separates Base Set,
Base Set 2, and Shadowless rather than returning every Charizard printing. Every result
carries a set field and a product_id you can pass straight to the other tools.
---
📑 Table of Contents
- Connect over MCP
- Why This Exists
- Data Coverage
- Quick Start
- Tools
- search_cards
- get_price
- get_merkle_proof
- oracle_status
- get_forecast
- simulate_price
- get_market_snapshot
- Architecture
- Configuration
- On-Chain Contracts
- License & Commercial Use
- Links
---
Why This Exists
AI agents are making decisions with market data — but how do they know the data is real?
Regular APIs require **trust**. You call an endpoint, you get a number, and you hope it's accurate. There's no way to verify it. For AI agents managing portfolios, executing trades, or assessing collateral, this is a problem.
**This MCP server solves it.** Every actively-priced product in the oracle is committed to a Merkle root on-chain daily — the USD root last advanced 2026-09-07 (feed frozen) and stays verifiable; Japanese, graded and sports panels advance daily. Any agent can request a Merkle proof for any card and independently verify the price against the LitVM LiteForge blockchain — no trust required.
What Makes This Different
| Feature | Regular Price API | LitVM TCG Oracle |
|---------|------------------|-----------------|
| **Data source** | Opaque server | 13.5M+ verified market observations |
| **Verification** | Trust the server | Merkle proof → on-chain verification |
| **Forecasting** | None | Calibrated conformal risk forecast — honest VaR |
| **Coverage** | Limited | 455K products, 290K actively priced |
| **For AI agents** | Manual integration | MCP — works in Claude, GPT, Cursor |
| **Blockchain** | None | LitVM LiteForge |
---
Data Coverage
The oracle indexes the full TCGPlayer catalog and tracks live market prices:
| Metric | Count | Description |
|--------|-------|-------------|
| **Total catalog** | 446,694 | All products across 25+ games and 85 categories |
| **Actively priced** | 276,788 | Products with a current market_price > 0 |
| **Price observations** | 13.5M+ | Daily snapshots collected over months |
| **Merkle-provable** | 276,788 | Only actively-priced products are committed on-chain |
| **Zero-price entries** | ~157K | Tokens, promos, bundles, foreign-market-only — searchable but not provable |
> **Transparency note:** Not every product in the catalog has a market price. ~157K entries are catalog metadata with no trading activity (token cards, unopened case listings, foreign-language promos, etc.). These are returned by search_cards but will return a 404 from get_merkle_proof because zero-price products are not committed to the Merkle tree. This is by design — you wouldn't commit unverifiable data on-chain.