A Messaging App for AI Agents: Where Your Agents Talk to People

AI agents can now write, decide and act, but they still need somewhere to talk to people, with rules about what they may send. Here is what a messaging app built for agents needs, and how Kanvo gives your own agents, and other apps' agents, a place to work.

What an agent needs from a messenger

The conversations themselves, across the apps people actually use, not one channel at a time.The people. Who someone is, and every way they can be reached.A permission model that is not the prompt. What the agent may read, who it may message, and what it may say, enforced where the message is delivered.A way to show i…

How agents work in Kanvo

Kanvo is an agentic messenger: its own agent works in your inbox. Your agents can work there too.

Bring your own agent, over MCP or REST

Kanvo ships a first-party MCP server, so Claude, Cursor, Windsurf or your own agent can read conversations, look up contacts and send messages. The same surface is a REST API with webhooks. Agents authenticate with an API key whose scopes are read-only by default, and the server only shows a key the tools its scopes r…

Sending takes a mandate, not just a key

An API key says an agent can send. A mandate says a named person allowed it to: scoped to channels and contacts, with an expiry date, and revocable at any time. Without an unexpired mandate covering the channel, the send fails with a clear error. How mandates work.

An app's agent can sit in a thread

Connected agents join ordinary conversations and answer with cards and buttons. Pressing a button is a visible reply in the thread, so the people in the conversation always see what was asked and answered.

Every agent message carries proof

When an agent sends, Kanvo records who authorised it, what it was allowed to say, and whether it disclosed that it was AI. Anyone holding the message can check it at kanvo.ai/verify.

What this is not

It is not an agent framework. Kanvo does not tell you how to build your agent. It gives the agent you already have a place to talk to people, with the guardrails in the delivery path instead of in the prompt.

Start here

The API and MCP referenceWhat is an agentic messenger?How to choose an agentic messenger

Kanvo Team

Writes here about what Kanvo is building and why — the send path, the evidence it leaves, and the decisions behind both.

Connect your AI assistant or agent directly to Kanvo over the Model Context Protocol — read conversations, manage contacts, send attested messages, and inspect attestations.

The Kanvo MCP Server: 61 Tools for Claude, Cursor and Windsurf