We Used Crypto.com's Public MCP: A Safe Read-Only Test

On September 1, 2026, WagerX connected directly to Crypto.com's official public Market Data MCP. We initialized the protocol, inspected its complete tool registry and executed two representative read-only calls. No account, API key, exchange credential or payment was involved.

200HTTP response

9read-only tools listed

2/2sample calls succeeded

What we tested

The endpoint was https://mcp.crypto.com/market-data/mcp. WagerX sent an MCP initialize request using streamable HTTP. The server responded with HTTP 200, identified itself as Crypto.com Exchange MCP version 0.1.0, and negotiated protocol version 2025-03-26.

We then requested tools/list. The server returned nine tools:

get_candlestick
get_ticker
get_index_price
get_trades
get_instruments
get_tickers
get_mark_price
get_instrument
get_book

Every listed tool carried readOnlyHint: true and destructiveHint: false. An annotation is useful machine-readable intent, though it should not replace permission controls or independent testing.

Test call 1: BTC/USDT ticker

We called get_ticker with instrument_name set to BTC_USDT. The request returned both human-readable text and structured content containing the instrument, last price, 24-hour high and low, best bid and ask, sizes, volume and a UTC timestamp.

{
  "instrument_name": "BTC_USDT",
  "last": "77984.72",
  "best_bid": "77984.72",
  "best_ask": "77984.73",
  "timestamp": "2026-09-01T09:42:13.956Z"
}

These values are a dated test receipt, not a current quote. Crypto markets move continuously.

Test call 2: three-level order book

We called get_book for BTC_USDT with a depth of three. It returned three bid levels, three ask levels and a timestamp. The best bid was 77984.72 and the best ask was 77984.73 at 2026-09-01T09:42:16Z.

The structured response is particularly useful for agents because a client does not need to extract prices from prose. It can validate named fields against the published output schema.

What this proves

The official endpoint was publicly reachable from WagerX's test environment.
MCP initialization and tool discovery worked without an account or key.
Nine tools were exposed with read-only, non-destructive annotations.
Representative ticker and order-book calls returned successful structured data.

What this does not prove

We did not exhaustively execute all nine tools or all parameter combinations.
We did not benchmark long-term uptime, latency, rate limits or data completeness.
We did not compare the returned values against an independent market-data source.
We did not test trading, balances or account permissions; those capabilities were not present in the observed registry.
The test is not financial advice, an exchange endorsement or a security certification.

Why narrow can be better for agents

This MCP does one bounded job: public market-data retrieval. That narrow scope makes the permission model easy to explain to a newcomer and reduces the consequences of a mistaken tool choice. In contrast, an MCP that mixes public data, balances, transfers and trading needs stronger authorization, confirmation and audit controls.

For professional developers, the combination of JSON schemas, structured output and tool annotations makes integration straightforward. Production clients should still validate every response, handle stale or missing data and impose their own timeouts and limits.

WagerX verdict

Observed working for representative public read-only retrieval. The evidence level is stronger than documentation-only discovery or a handshake-only check, but narrower than an exhaustive reliability assessment.

See the maintained Crypto.com Market Data MCP evidence record or compare it with other products in the Crypto Exchanges category.