Messaging and voice for AI agents.

Pick the way in

Agents usually need two things: facts to plan with, and a way to act. BroadNet keeps them apart. The knowledge server answers questions without an account; sending always needs your token.

What an agent can send

The same rules apply whichever surface the agent uses: a registered sender, a consented recipient and content that fits the channel.

A worked example: an agent plans an OTP launch

A user asks an agent to plan one-time passcodes to the UAE. The agent calls four tools on the public knowledge server. The results below are the server's real output, trimmed to the fields the agent uses.

Copy and paste

The examples use fictional numbers only. Read the token from an environment variable; never paste it into a chat or a committed file.

{
  "mcpServers": {
    "broadnet-knowledge": {
      "type": "http",
      "url": "https://www.broadnet.me/mcp"
    },
    "broadnet": {
      "type": "http",
      "url": "https://mcp.broadnet.me/mcp",
      "headers": {
        "Authorization": "Bearer ${BROADNET_TOKEN}"
      }
    }
  }
}

curl -X POST https://www.broadnet.me/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "MCP-Protocol-Version: 2025-06-18" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call",
       "params":{"name":"estimate_price",
                 "arguments":{"country":"UAE","channel":"sms","volume":50000}}}'

curl -X POST https://api.broadnet.me/sms/sendSms \
  -H "Authorization: Bearer $BROADNET_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "senderId": "BroadNet",
    "destinationNumber": "447700900123",
    "message": "Your verification code is 482913",
    "messageType": "TEXT"
  }'

# 200 OK → response body is the message ID, e.g. 2603121407537573731

// npm install @broadnet/cpaas-sdk axios
import { Configuration, SMSSMSApi } from "@broadnet/cpaas-sdk";

const sms = new SMSSMSApi(
  new Configuration({
    basePath: "https://api.broadnet.me",
    accessToken: process.env.BROADNET_TOKEN!,
  }),
);

const res = await sms.smsSendsmsPost({
  sMSSmsRequest: {
    senderId: "BroadNet",
    destinationNumber: "447700900123",
    message: "Your verification code is 482913",
    messageType: "TEXT",
  },
});

console.log(res.data); // message ID

# Send an approved WhatsApp template
curl -X POST https://api.broadnet.me/whatsapp/message/sendbulk \
  -H "Authorization: Bearer $BROADNET_TOKEN" \
  -F batchId=0 \
  -F campaignName=order-updates \
  -F destinationNo=447700900123 \
  -F messageType=TEMPLATE \
  -F templateName=order_update \
  -F templateLanguage=en_US \
  -F bodyTemplateVariables=INV-400

# 200 OK → the response carries the batchId; per-recipient status
# (SUBMITTED → SENT → DELIVERED → READ) arrives on your webhook.

Machine-readable entry points

Agents that browse rather than connect can read everything from plain files. None of them need a login.

• llms.txt
  /llms.txt

  The site index for language models, with a section on agent access and constraints, plus one file per language.

• Agent skills
  /.well-known/agent-skills/index.json

  Five SKILL.md runbooks: coverage and price, sender-ID registration, a first test SMS, test access, and connecting to the MCP servers.

• A2A agent card
  /.well-known/agent-card.json

  An Agent2Agent card for the same read-only knowledge tools, served over JSON-RPC at /a2a.

• API catalog
  /.well-known/api-catalog

  An RFC 9727 linkset pointing to the OpenAPI spec, the API docs, both MCP servers and the status page.

• MCP server card
  /.well-known/mcp.json

  The server card describing both MCP servers, their transport and authentication.

• OpenAPI 3.1
  /developers/openapi.yaml

  The full REST API description: download it, or generate a typed client from it.

• Markdown twins
  /pricing/sms/uae.md

  Add .md to most page URLs, or send Accept: text/markdown, for a clean Markdown copy of the page.

• Data files
  /data/coverage.json

  Coverage, published prices and regulations for every market as JSON (CSV for coverage and pricing).

• Status JSON
  /status/api/v2/summary.json

  Measured reachability of the website, API gateway and both MCP endpoints, in the Statuspage v2 summary shape.

Rules an agent must follow

An agent that sends messages acts for your brand with your credentials. Build these rules into its instructions and its tools.

• Consent before content

  Send marketing only to recipients who opted in, and keep the opt-in record. The sender_id_rules tool returns each market's opt-in rules.

• Honour opt-outs

  Promotional messages carry the market's opt-out keyword, such as STOP. An agent never messages a number that opted out.

• Registered sender IDs only

  Many markets require pre-registration of alphanumeric sender IDs. The agent uses the sender IDs approved on your account, never an invented one.

• Respect quiet hours

  Several markets bar marketing at night. sender_id_rules returns the sourced quiet hours; OTP and alerts are usually allowed around the clock.

• Guard OTP flows against pumping

  Bots can trigger OTP sends to numbers they profit from. Rate-limit code requests per number, IP and device, and block destinations you do not serve. MyOTP ships these protections on by default.

• A person confirms every send

  Have the client ask before any tool that sends, especially to many recipients, and enable only the tools the agent needs.

• Test on your own number

  With test credentials, send only to your own verified number. Examples and prompts use fictional numbers.

• Keep the token out of the chat

  The token is a send credential. Load it from an environment variable, never from a prompt, and revoke it if it leaks.

Related

Questions