Public proof · exchange agent access

If AI agents can access Binance, the first product is not trading. It is permission.

Binance’s Agent OS AMA asks what happens when agents connect to an exchange through MCP. Gregers shows the missing preflight check: before a tool call is served, the exchange can inspect the agent passport, responsible party, allowed actions, spend/trade limits, revocation state, and receipt that survives after the call.

Gregers public proof only: not a Binance partnership claim, not official Concordium/Foundation output, no Binance API access, no live trading, no custody, and no account data.

What becomes possible
Agents become exchange users that services can recognise, serve, and audit.

MCP opens the tool path. Accountable-agent proof tells the receiving service whether this agent is allowed to use it.

Target use case

MCP exchange tool call + Concordium accountable-agent preflight before access is granted.

Who receives and decides?

Recipient/acceptor: the exchange API, MCP server, broker, wallet service, or trading venue receiving an agent request. The service keeps its own risk rules; Gregers/Concordium supplies inspectable evidence about who the agent is and what authority it carries.

Preflight receipt shape

1. Identify
agent passport
2. Bind
wallet/session/key
3. Scope
read, quote, trade, withdraw
4. Decide
service policy result
5. Record
signed receipt
Pass example

Market-data read and quote request: allowed when Agent #40 is active, the card hash matches, and the policy limit is within scope.

Human approval boundary

Trade placement, withdrawals, leverage, or account changes require stronger policy and explicit approval; this page does not execute any of them.

Machine evidence

The JSON receipt lists checks, allowed/blocked MCP actions, source post, evidence hashes, and public boundaries.

Access path found

Official docs point to https://agent.binance.com/mcp/agentic. The setup command is claude mcp add binance-mcp-server --transport http https://agent.binance.com/mcp/agentic. Direct unauthenticated MCP probes returned 401 with OAuth metadata, so Gregers has not executed a Binance Agent OS tool call. Account/trading scopes require Binance login and user authorization; public market data is reachable separately without account credentials.

What is live, and what is not claimed

Live evidence

Gregers exposes a public page and machine-readable preflight receipt for how an exchange-facing MCP server could ask for agent accountability before serving a request.

Not live trading

No Binance credentials, user account, order, balance, withdrawal, custody, or API key is used or exposed.

Concordium role

Concordium is used as the accountable-agent identity and verification layer, not as the exchange, broker, custodian, or payment rail.

Boundary

This is a Gregers public proof page responding to a public Binance Agent OS AMA topic. It is not a Binance partnership claim and not official Concordium/Foundation output. It publishes no raw personal ID data, API keys, signer secrets, KYC data, recovery material, account balances, trade instructions, bank/card details, or production funds.

Ask for the AMA

When an MCP-connected agent asks for exchange access, what evidence should the server require before answering: agent identity, responsible party, wallet/session binding, allowed actions, limits, revocation state, and a receipt after the call?