Governed agent wallets for DeFi risk curators
Your curator agent recommends a parameter change. Broadcast stays with the curator team. Every recommendation is a structured, timestamped payload for your review queue.
{ "mcpServers": { "walletsuite": { "command": "walletsuite-mcp", "env": { "WALLETSUITE_API_KEY": "$WALLETSUITE_API_KEY", "MCP_BANDS": "read,prepare" } } }}Closing the gap between recommendation and broadcast
Risk curators run agent-driven parameter management for lending protocols and vaults. The model reads chain state, runs stress scenarios, and recommends an adjustment. The gap is between the model saying retune and the parameter change landing on-chain. Today that gap is a human manually pushing a multisig transaction with no structured handoff from the model.
Band filtering closes the handoff without moving broadcast to the model. A curator agent configured with read and prepare can query chain state, construct the parameter-change transaction, and sign nothing. The prepared payload lands in the curator team’s internal review queue. A human curator reviews, approves, and co-signs through the existing multisig.
Every prepared payload is a structured JSON artifact — chain, parameters, and policy evaluation — ready to drop into your internal review queue. The hash-chained audit trail on your infra begins once the curator team co-signs through their multisig; DAO post-hoc review stops being a chat-log reconstruction exercise.
One band cap, no signing surface
The curator agent runs with MCP_BANDS=read,prepare— the tool registry at startup contains research and unsigned-tx construction only. OWS is off on this instance: there is no signing surface in the agent’s session. The prepared transaction is emitted as structured JSON, ready for a human curator to co-sign via your existing multisig signer.
The curator agent cannot broadcast a parameter change because the tool to do so does not exist in its session. Every prepared transaction is a structured payload — chain, timestamp, policy evaluation — ready for your internal review queue. The hash-chained audit trail starts when the curator team co-signs through their multisig.
Three properties that drop out of the pattern.
- 01Every model-recommended parameter change is a structured, auditable payload before a human signs.
- 02DAO post-hoc review is a clean handoff — structured prepare payload from the model, hash-chained JSONL once the multisig co-signs. No chat-log reconstruction.
- 03The model’s world ends at prepare. Broadcast stays with your curator team’s existing multisig signer.
Now reviewing a limited number of design partners.
See all use cases