AI agents

Notix is built to be used by AI agents as well as by people. This page gathers what you need to let an agent send, read and act for you, within limits you set.

Connect an AI app

Connect your AI gives Claude, ChatGPT, Cursor, Copilot or Windsurf a key in three steps. The key sends from your own verified domain, within limits you set.

The MCP server

Notix hosts an MCP server at https://app.usenotix.dev/api/mcp. Send your key in the Authorization: Bearer header and the app gets only the tools that key may use. To run the server on your own computer instead, use the notix-mcp package. It reads your key from NOTIX_API_KEY.

NOTIX_API_KEY="notix_..." npx -y notix-mcp

Test with a dry run

Ask your agent to pass dry_run: true on a send. Notix answers with who it would reach, what it would cost and whether it would be allowed, and sends nothing. Over the API, the same check is the Notix-Dry-Run: true header. See Test emails.

Keep an agent inside its limits

• A Live AI key sends from one verified domain only.
• Ask me first makes a person approve any send to more than a set number of people.
• A Daily limit stops the key for the day.
• The audit log records every change an agent makes, every tool call and every refusal.

Details are in API keys and permissions.

Received mail is data, not orders

Replies, contact fields and imported CSV rows are written by other people. Notix labels them as untrusted text for your agent. Read AI agents and untrusted email.

Docs your agent can read

• /docs/llms.txt: a short map of every docs page.
• /docs/llms-full.txt: every docs page in one Markdown file.
• Each page has Copy page as Markdown, Open in Claude and Open in ChatGPT, so you can hand a page to an assistant in one click.
• The API reference is built from the same OpenAPI document the API serves.