Station 0 · never used an agent · read-only
First Contact
Runs today
Ten minutes from now, your own AI reads live Gryps market data and tells you what it can and cannot do. No account, no wallet, no money.
Everything that comes later rests on you seeing what your agent sees. That starts with live venue data in a chat you control. The connection is read-only by architecture, not by promise.
The prompt
You are connected to Gryps in read-only mode. Do not do anything yet. Show me around first. 1. Run gryps_venue_status. Tell me whether the venue is healthy, which chain and contract settle it, and which sources you read. Read the limitations section out loud. 2. Run gryps_list_markets. Tell me how many markets the engine lists, and read me the limitation about what a listed market does and does not prove. 3. Run gryps_get_fee_schedule, then gryps_friction_floor for BTC. Tell me the fee ladder, the number a trade has to beat, whether that number is a lower bound, and why. 4. If any number is missing or a source is unreachable, say exactly that. Do not fill gaps with numbers you remember from training. Explain every number you show me. If something surprises you, say so.
Runs as written with gryps-agent-mcp, installed in one line. Read-only.
If any number is missing, your agent should say exactly that. An agent that fills gaps with remembered numbers is the thing this whole library trains out.
What good looks like
Your AI reads back the venue's health and settlement chain, the engine's market count with its own caveat attached, a live 10-tier fee ladder, and the number a trade has to beat with the word "lower bound" next to it. You should come away thinking "it is reading the actual venue", because it is, and when a source is unreachable it says so instead of improvising.