frame: trust-and-outcome — before an agent acts it verifies the counterparty; after it acts it proves the result products: - id: trustscan · slot: trust · direction: outward · mcp: /mcp/trustscan (4 tools) · live - id: agent-watch · slot: trust · direction: outward · mcp: /mcp/agent-watch (8 tools) · status: live (gateway-verified 2026-10-07) - id: perimeter-watch · slot: trust · direction: outward · mcp: /mcp/perimeter-watch (6 tools) · free_snapshot: https://entradox.github.io/perimeter-watch-site/ - id: benefits-city · slot: outcome · direction: outward · mcp: /benefits/mcp (6 tools · verified 2026-10-13) - id: cited · slot: outcome · direction: outward · mcp: /mcp/cited (9 tools) · free_scan: https://entradox.github.io/cited-site/ - id: agent-ledger · slot: outcome · direction: spine · mcp: /mcp/ (18 tools) · rest: /v1/* pricing: {free: "≤3 agents", pro_pass: "$0.10 USDC / 24h via x402, renewable — the only paid path"} - id: workbench-kit · slot: trust+outcome · direction: inward · kind: local kit (no MCP — gates, cards + receipts run on the operator's machine) · install: npx @aiagentscity/setup workbench · verify: wb verify <receipt_id> index: Accept: application/json on / returns this map as JSON · /v1/products · /.well-known/mcp.json auth: workspace_key (human) | X-PAYMENT x402 (agent, no human)
TRUST products verify the counterparty before an agent acts; OUTCOME products prove the result after. Each is live, independently usable, and agent-callable — MCP on the six services; Workbench Kit installs on the operator's machine over npx.
Your agents spend money. Give each one a spending limit. Meters every call, shows a live P&L per agent, and refuses the call that would cross the budget — before the provider is contacted.
HTTP/1.1 402 Payment Required {"error":{"type":"budget_error", "message":"blocked before the provider"}}
Free: 3 agents, every rail, enforced caps. Paid: $0.10 USDC = 24h Pro over x402, renewable — the only paid path, bought by the agent itself.
Non-human traffic is hitting your APIs and you can't see it. What changed this week, alerts the moment a new agent calls your endpoints, anomaly summaries — before they become incidents or invoices.
"Alert me when a new agent calls our API, with a weekly activity summary."
Status: live — MCP handshake and all 8 tools verified working (2026-09-20). Tool calls execute through the city gateway, which translates them to the product's documented REST API. Live status →
Attackers view your endpoints from the outside. So do we. Dangling DNS, expiring certs, lookalike domains — scanned passively, briefed weekly.
"Scan example.com for dangling DNS and cert expiry, then check lookalikes."
Status: live — MCP handshake and all 6 tools verified working (2026-09-20). Tool calls execute through the city gateway, which translates them to the product's documented REST API. The free browser snapshot works today. Live status →
Your agent is about to install a stranger's MCP server. TrustScan answers the question every agent should ask first: is this safe to wire in? Invisible-Unicode prompt-injection, dangerous code patterns (MCP001–MCP006), hardcoded secrets, typosquat names — a 0–100 score, a letter grade, and evidence. Live over MCP today, read-only, no signup.
Your customers stopped Googling. They ask AI. A scan of what the AI engines actually say about a business — with verbatim evidence, not vibes. Free.
"Scan Gentry Dentistry of Suwanee for AI visibility — verbatim quotes."
Status: live — MCP handshake and all 9 tools verified working (2026-09-20). Tool calls execute through the city gateway, which translates them to the product's documented REST API. The free browser scan works today. Live status →
Bank and card bonuses have fine print that decides whether they pay. Hand-verified offer records — exact requirements, minimums, deadlines, source URLs — so an agent can check the outcome before it promises the result.
"Which checking-account bonuses expire this month, and what does each actually require?"
Status: live — MCP initialize, tools/list and tools/call verified (2026-10-13). search_bonuses and expiring_soon return verified records unauthenticated — the data is the free proof. Live status →
Anyone can say the agent did the work. Prove it. One command installs the discipline on the operator's machine: every task gets a frame card, every write is gated against the card's scope, and every close produces a receipt anyone can recompute against the ledger.
npx @aiagentscity/setup workbench
A local kit, not a service — no MCP, no account, nothing leaves the machine unasked. Wires into Claude Code, Codex, Cursor, OpenCode, Hermes and OpenClaw. Free: the whole kit — the city does not sell a human tier for it.