Every product answers one of two questions: TRUST — can it trust the counterparty, before it acts. OUTCOME — did it work, after. Seven products, one MCP catalog, zero human required. Read the contract, connect, transact.
the agent-native surface, playing next to the contract — the whole site works like this
the installer detects claude-code, codex, cursor, opencode, hermes + openclaw and registers the right way for each. manual fallback: claude mcp add --transport http agent-ledger https://aiagentscity.com/mcp/
$0.10 · zero clicks.
POST /v1/billing/x402 + X-PAYMENT header → 24h AgentLedger Pro on the workspace your wallet resolves to. No signup flow, no card form, no human.
TRUST products verify the counterparty before an agent acts; OUTCOME products prove the result after. Six are MCP-native — tool counts are the live tools/list output, not marketing. Workbench Kit is a local kit installed over npx, not an MCP endpoint.
Per-agent spend caps with 402 enforcement before the provider is ever called. The call that would break the budget never reaches the provider.
Monitoring for the agent economy. Know the moment a new agent touches your API.
Is your endpoint still what it claims — checked from the outside? Cert expiry, dangling DNS, lookalikes — caught before they're incidents.
Scan before you trust. Know who you're dealing with before your agent does.
Verify the claim that AI recommends this brand — verbatim evidence.
Will this action actually produce the payout? Hand-verified offer records — requirements, deadlines, source URLs — an agent can check before it promises the result.
The audit trail for agent work. Frame every task, gate every write against its scope, close with a receipt anyone can recompute.
They spend money. They call your APIs. They represent businesses. AI Agent City is the operations layer they run on — budgets, monitoring, security, trust, and discovery. Live software, not slides.
Machine-to-machine payments settle on-chain for fractions of a cent. The moment agents spend, someone has to set the budget.
Non-human traffic already hits production endpoints, and most teams can't see it. You can't secure or bill what you can't attribute.
Customers ask ChatGPT and Claude instead of Googling. If the models don't recommend you — with receipts — you're invisible.
Each answers one question end to end: can it trust the counterparty, or did it work. Each is live and independently usable.
Per-agent spend caps with 402 enforcement before the provider is ever called. The call that would break the budget never reaches the provider.
Monitoring for the agent economy. Know the moment a new agent touches your API.
Is your endpoint still what it claims — checked from the outside? Cert expiry, dangling DNS, lookalikes — caught before they're incidents.
Scan before you trust. Know who you're dealing with before your agent does.
Verify the claim that AI recommends this brand — verbatim evidence.
Will this action actually produce the payout? Hand-verified offer records — requirements, deadlines, source URLs — checked before your agent promises the result.
The audit trail for agent work. Frame every task, gate every write against its scope, close with a receipt anyone can recompute.
An agent bought 24h of AgentLedger Pro over x402 on Base mainnet — real USDC, no human in the loop.
Trace viewers report after you've paid. AgentLedger refuses the over-budget call before the provider sees it.
Flat-rate token pricing was 7.4× wrong on a real session. AgentLedger prices cache-aware.
The changelog is the proof the stack is alive — every entry is a shipped outcome, not a progress update.