Give your AI agent a Canadian messaging API

• Send SMS, list messages and read delivery status
• Start a verification code, check it and list attempts
• Search, buy and release Canadian phone numbers
• Record and check CASL consent, and handle opt-outs
• Manage webhooks and replay failed deliveries
• Read account details, usage and the daily sending limit, and manage API keys
• Send email, schedule sends, and manage domains, suppressions and templates

Five guided prompts walk an agent through common jobs: a compliant send, buying a Canadian number, setting up webhooks, a compliance audit and an erasure request.

Then ask in plain language

• “Find an available 416 number, tell me what it costs, and don't buy anything yet.”
• “Check whether this customer has CASL consent on file before we text them.”
• “Set up a webhook for delivery receipts and replay any deliveries that failed.”

The hosted endpoint needs nothing installed. Add it by URL and sign in: a page on honkio.ca asks you to approve the connection and choose test or live mode. You can send an API key in a header instead.

claude mcp add --transport http honkio https://mcp.honkio.ca/mcp

Prefer to run it yourself? The same server is on npm as @honkio/mcp, and npx fetches it on first use.

{
  "mcpServers": {
    "honkio": {
      "command": "npx",
      "args": ["-y", "@honkio/mcp"],
      "env": { "HONKIO_API_KEY": "mk_test_YOUR_KEY" }
    }
  }
}

It works with Claude Code, Claude Desktop, claude.ai connectors, the Claude Messages API, Cursor, VS Code and any other client that speaks MCP over HTTP or stdio.

• Test keys simulate sends and verifications for free, and work before your first top-up.
• Signing in shows what the agent may do, and a live connection also asks for your account password. Each connection is listed with your API keys and can be revoked there.
• API keys carry permissions per resource, so you can give an agent a key that reads messages but cannot buy numbers.
• Buying or releasing numbers, changing webhooks and running erasure need a live key.
• Every tool says whether it only reads or can change or delete something, so clients that support it can ask you first.
• Text someone else wrote, such as an SMS sent to your number or a received email, comes back marked as untrusted content: the agent is told to read it as data and never to follow it as instructions. Reading received email stays off until you turn it on.
• An agent's sends go through the same API as your code's: the same CASL consent checks, STOP handling and sending limits apply.

llms.txt indexes the docs in the format language models read, llms-full.txt holds every endpoint, error code and limit in one file, and the OpenAPI document describes every request and response. Every error has a stable code and a message in English and French.

Actions are billed from your prepaid CAD balance at the usual prices: $0.03 per SMS segment, $0.25 per verification plus its SMS, and $2.50 a month for a local number. Nothing is billed in test mode.

Is the HonkIO MCP server free?

Yes. The server costs nothing to use. The messages, verifications and numbers an agent uses are billed at the usual prices, and nothing is billed in test mode.

Which AI clients does it work with?

Claude Code, Claude Desktop, claude.ai connectors, the Claude Messages API, Cursor and VS Code connect to the hosted endpoint by URL. Any other MCP client can use the hosted endpoint over HTTP, or run the npm package over stdio.

Can an agent spend money on my account?

Only on a live connection. Explore with a test key, where sends are simulated and nothing is charged. On a live key, texts and numbers cost real money, so give the agent a key with only the permissions it needs.

Can a text message trick my agent?

The server marks text written by someone else, such as an incoming SMS, as untrusted content, and tells the agent to treat it as data, never as instructions. Narrow key permissions are a second line of defence.

Where is my data stored?

In Canada (AWS ca-central-1, Montréal), as for every HonkIO request. Requests are processed on US-hosted infrastructure before they are stored. What you type into your AI client, and what the tools return to it, is handled by that client's provider under its own terms.

Does an agent have to follow CASL?

An agent uses the same API as your code, so the same rules apply: a commercial text needs CASL consent on file, a STOP reply opts the person out, and sending limits hold. A guided prompt checks consent before it sends.

What licence is the npm package under?

The npm package, @honkio/mcp, is published under the MIT licence.

• SMS API

  Send and receive SMS with Canadian local and toll-free numbers, with CASL consent and STOP replies enforced in the API.

• Verify

  One-time codes by text message, to confirm that a Canadian phone number belongs to your user.

• Email API

  Send transactional email from your own domain and receive replies over the API, with CASL checks and data stored in Canada.

• CASL for developers

  What Canada’s anti-spam law asks of the code that sends texts and email: consent, identification, unsubscribes and records.