Architecture

Narrative

OpenFi Lido MCP Suite is a fork-first, mainnet-ready Lido system that lets humans fund AI agents with wstETH-backed yield while structurally protecting principal, monitors Lido Earn vault positions in plain language, and exposes staking, treasury, and governance actions through a reusable MCP server.

System Diagram

flowchart LR
  U["User / OpenClaw ingress"] --> CP["OpenFi Control Plane"]
  CP --> CO["Coordinator"]
  CO --> LA["lido-agent (OpenFi adapter)"]

  subgraph CORE["Shared Lido Domain Core (packages/shared/src/lido)"]
    CL["Clients + networks"]
    AD["Adapters (staking, withdrawals, rewards, earn, governance)"]
    FM["Schemas + formatters"]
  end

  LA --> CORE
  MCP["Standalone MCP server (packages/lido-mcp)"] --> CORE
  OFMCP["OpenFi MCP server"] --> CORE
  MON["Monitor (packages/monitor/src/lido)"] --> CORE

  CORE --> EX["Execution chain (mainnet fork/default or mainnet live)"]
  LA --> TR["Sepolia trust/identity (ERC-8004)"]
  EX --> RC["Receipts + verification"]
  RC --> WEB["Web dashboard"]

Trust vs Execution Split

flowchart LR
  subgraph T["Trust Domain"]
    TS["Sepolia (11155111)"]
    ID["ERC-8004 identity + manifest/trust checks"]
  end

  subgraph E["Execution Domain"]
    FK["Mainnet fork (default)"]
    MN["Mainnet live (optional)"]
    LD["Lido protocol + YieldSafeTreasury + ComputeCreditsMerchant"]
  end

  TS --> ID
  FK --> LD
  MN --> LD

Lifecycle

OpenFi run lifecycle for Lido actions:

1. discover
2. plan
3. trust check (Sepolia)
4. policy check
5. dry run
6. execute
7. verify
8. summarize
9. record

Invariants

• Agent cannot withdraw principal from YieldSafeTreasury
• Principal baseline tracked in stETH-equivalent units
• Spend path rechecks principal floor post-transfer
• Whitelist/per-tx/window/sub-agent budget checks enforced onchain
• Receipts always include chain/domain labels

Threat Model

• Compromised agent key
• Malicious merchant recipient
• Stale APR/health data
• RPC/API downtime
• Owner misconfiguration (whitelist/caps/budgets)
• Rebase/accounting misunderstandings
• Trust/execution chain confusion

Security Assumptions

• Owner key custody is secure
• Sepolia trust metadata is correctly registered
• Execution signer has expected permissions
• Lido addresses are sourced from official docs only