All guides
AI agents & MCP
The whole CRM is exposed as a Model Context Protocol server, so agents and services can operate it directly.
Integrationsabout 5 min
By the end you'll be able to
- Connect Claude, ChatGPT, Codex or Claude Code to the CRM
- Have an agent design a whole email, images and all
- Turn a spreadsheet of send dates into a month of drafts
- Understand what a connected agent is allowed to do
Walk through it
0 of 6 steps done0%
Settings → AI agents holds your connection URL, step-by-step setup for each tool, and the live list of everything an agent can call. Keep it open in a tab while you set your client up.
Good to know
- A connected agent is you. It sees exactly the brands you can see and can do exactly what you can do — no more. Someone else connecting their own client gets their own access, not yours.
- Emails an agent writes still go through approval before anything is sent, and consent and suppression rules still apply. Deletes are soft deletes, so a mistake is recoverable.
- A date an agent sets on a draft is a plan, not a commitment — nothing sends off it. Approving the email is still a separate, human step, so a forty-row spreadsheet cannot turn into forty sends by itself.
- Agents build emails from the same blocks the editor uses, so anything they make can be opened and edited by hand afterwards. They cannot hand-write raw HTML into an email.
- Claude on the web and ChatGPT connect from Anthropic’s and OpenAI’s servers, not from your computer. They need Zenia deployed at a public address — they cannot reach a development copy running on your laptop.
- Automations that cannot open a browser use a service key instead, issued by whoever runs the API service. There is no self-serve key generation, on purpose.
Common questions