Six tools over the Model Context Protocol: live markets with fees-as-data, per-market detail, funding parameters, open interest, chain-side venue liveness, and operational quirks lookup. Point Claude Desktop — or any MCP client — at it and your agent reads Gryps.
§1Capabilities
Serves live market state — symbols, leverage caps, and fee schedules read from the venue registry at call time, never hardcoded. When pricing changes, your agent sees a data change, not a broken assumption.
Answers venue-liveness questions from the chain itself: state-root cadence, sequence number, and custody balance via plain eth_call. This layer works even when the venue API is unreachable — it reads the contract, which cannot have its DNS deleted.
Exposes the Quirks Registry as a queryable tool, so the operational facts an agent must respect (scaling conventions, funding grids, accounting identities) arrive as structured rules.
§2Design
MCP-native. Standard Model Context Protocol server. Works with Claude Desktop, Claude Code, and any MCP-compatible client — one config entry and the tools appear.
Two truth sources. API-layer reads for market data; chain-layer reads for liveness. Each tool declares which layer it reads from, so your agent can reason about trust and staleness explicitly.
Read-mode first. The open build is deliberately read-only. Write mode ships behind the same adapter seam once the v2 engine write path is stable — no re-integration required.
§3Availability by tier
| Open | Read-mode server, full source, non-commercial license. |
| Pro | Write mode plus a hosted endpoint with SLA. |
| Institutional | Dedicated instance, custom tools, commercial license. |
Indicative until launch.
§4 · Stated plainly
Read-mode only today; order submission is gated until the v2 engine API stabilizes. Chain-side tools cover the Polygon deployment.