Cross-agent settlement & metering for multi-agent workflows

When several agents collaborate on one task, someone has to keep track of who did what and who owes whom. JiZhiMCP ships a neutral, tamper-evident settlement ledger for exactly that.

The problem

Multi-agent systems split work across specialized agents. Without a shared ledger you can't reliably answer: which agent did the work, how much resource did each consume, and how should the reward be split. Building that accounting yourself means accounts, transfers, idempotency, and audit trails — a lot of non-trivial code.

The five tools

ToolWhat it does
settlement.accountOpen or query a settlement account with a credits balance
settlement.transferAtomic credit transfer between two agents — idempotent, no overdraft, hash-chained
settlement.ledgerQuery an account's balance and transaction history for reconciliation
audit.recordImmutable audit trail with SHA-256 digests
metering.recordPer-agent usage metering and aggregation

Why it's safe

Example

An orchestrator settles 5 credits to agent-a for a completed sub-task:

{
  "method": "tools/call",
  "params": {
    "name": "settlement.transfer",
    "arguments": {
      "from_account": "orchestrator",
      "to_account": "agent-a",
      "amount": 5,
      "reason": "task.complete",
      "idempotency_key": "settle-001"
    }
  }
}

This makes JiZhiMCP a neutral settlement layer for multi-agent systems — a distinct capability you won't find in a plain MCP gateway or a directory listing.

What is an MCP gateway? · Host an API as MCP · GitHub