The agent-callable layer for SMB transactions.

Agent Broker is the MCP server that lets autonomous AI agents (Claude, Cursor, Continue, any x402-aware client) actually do business with the long tail of small and mid-sized businesses worldwide — finding them, verifying them, booking appointments, sending messages, escalating to a human when stuck — with full TCPA / GDPR / CASL / PDPL compliance enforced at runtime by a non-bypassable gate.

Browse the live API → Connect to Claude Pricing

Example: an agent gets a real consumer request —
"Book me a haircut at https://cal.com/jane-salon next Tuesday at 3pm"
Or an SMB asks its agent to text its opted-in customers about a sale. Or a customer texts the salon "STOP" — our handle_inbound classifies the opt-out and records it in the consent store automatically. The compliance gate decides what's allowed, not the marketing copy.

Five message types, four channels, 22 jurisdictions.

What we facilitate

Consumer-initiated bookings. SMB-initiated messages to opted-in customers (marketing, reminders, transactional). Voice calls with two-party recording consent. Cold-start discovery via import_booking_url. Inbound classification + automatic STOP / opt-out handling.

What the gate rejects

Marketing to recipients without a verified consent_record_id. Bulk / list-based / drip campaigns. Cold outreach to non-opted-in numbers. A/B test sends. Spam by any definition. The gate runs synchronously before every send and returns a structured compliance_violation receipt on rejection.

How enforcement works

/compliance/check runs before every outbound channel call. TCPA, GDPR, CASL, PDPL rules across 22 jurisdictions, including GCC (UAE, SA, OM, QA, KW, BH). A request that violates returns a structured receipt and never reaches a carrier.

Live activity

Public counters from this service. Update every 8 seconds. Numbers reset on each deploy.

0
Agent requests
0
Businesses returned
0
Messages sent
0
Operations completed

Connect in one line, in any agent ecosystem.

We expose the same 12 tools through every protocol agents speak today.

MCP — Claude Desktop / Cursor / Continue

{
  "mcpServers": {
    "agent-broker": {
      "url": "https://smb-broker.onrender.com/mcp",
      "headers": {
        "X-Agent-Identity": "$TOKEN"
      }
    }
  }
}

OpenAI function calling

tools = httpx.get(
  "https://smb-broker.onrender.com"
  "/.well-known/openai-tools.json"
).json()["tools"]

Anthropic tool_use

tools = httpx.get(
  "https://smb-broker.onrender.com"
  "/.well-known/anthropic-tools.json"
).json()["tools"]

12 operations. One contract. Worldwide.

Same OutcomeReceipt schema for every operation. Same compliance gate. Same idempotency contract. No surprises.

find_business

Search verified SMBs by vertical + location + capability. ~$0.01

verify_business

Confirm an SMB's capabilities before committing. ~$0.02

send_message

Consumer-initiated transactional message (SMS / email / voice premium). ~$0.05 base

capture_lead

Structured intake when a consumer asks to be contacted by an SMB. ~$0.05

schedule_appointment

Direct API or voice fallback. $0.25 attempt + $0.75 on confirmed booking

send_transactional_confirmation

Booking / receipt / reminder — reliable delivery, OTP-grade. ~$0.03

handle_inbound

Classify and route inbound messages an SMB received. ~$0.08

escalate_to_human

Hand a stuck operation to a human operator. ~$0.50

get_status / get_outcome

Async polling and final result retrieval. ~$0.001

preview_cost

±5% accurate cost estimate before committing. Free

self_test

Live capability probe / health check. Free

import_booking_url

Turn any supported booking URL into a callable SMB. ~$0.005

What's actually here — verifiable, not vibes.

Every number on this row maps to something you can confirm with one curl. Nothing simulated, nothing aspirational — just what the live service does today.

13
Callable tools
12
Booking platforms supported
22
Jurisdictions with native compliance
2
Payment rails (x402 + Polar)
7
Discovery protocols
<50ms
p50 latency, edge-served globally
$0
Free tier · reads always free
~$0.001
Base-chain gas per agent call

Verify each: /.well-known/mcp.json for tools, /supply/platforms for the 12 booking integrations, /compliance/jurisdictions for the 22 rule sets, /manifest for the canonical contract, /health for live status.

Built right.

Discovery

7 agent protocols

MCP, OpenAI plugin, OpenAI tools, Anthropic tools, A2A, llms.txt, OpenAPI.

Compliance

Non-bypassable gate

Pre-check is the only path to outbound. PII stored as SHA-256 hash only.

Reliability

Fallback chain

direct_api → voice_ai → sms → email → web_form. Circuit breakers per channel.

Idempotency

24h TTL

Scoped per (agent_id, operation, key). Safe to retry.

Async

Webhook callbacks

HMAC-SHA256 signed. Up to 24h retry with exponential backoff.

Worldwide

Jurisdiction-detected

22 countries with native rules. International conservative default for the rest.