What MCP is
MCP lets AI assistants like Claude or ChatGPT work in your Luminobot account safely, with the access you choose.
MCP (Model Context Protocol) is a common language that AI assistants use to work with other apps. It lets AI assistants like Claude or ChatGPT work in your Luminobot account safely: you connect the assistant once, and from then on you can simply ask it to do things.
"Go through my unread messages, label them and draft replies."
"Plan three posts for each of our accounts next week at the best times."
"How did our Instagram do last week compared with the week before?"
The assistant uses Luminobot's tools — small, well-defined actions such as "list conversations", "save a draft post" or "read analytics" — to do the work, and tells you what it did.
What an AI assistant can do
- Inbox: read conversations, suggest and write replies, label, assign, snooze, add internal notes, route to a team, use your saved replies and macros.
- Publishing: draft, schedule and cancel posts, write captions in your brand voice, find the best times to post, turn a blog post into drafts for every network, upload media.
- Contacts: search contacts, update details, see segments.
- Workflows: see your workflows and their runs, run a workflow, publish or restore versions.
- Analytics: read account and post results, AI insights, competitors and ad results.
- Approvals: see and decide AI actions, workflow approvals and post approvals.
The full, always up-to-date list is on MCP tools.
It acts as you
The assistant always works as the person who connected it. It can never do more than you can do yourself, and every request re-checks your current access. If your role changes or you leave a workspace, the assistant's access changes with it straight away.
When you connect it, you also choose:
- which workspaces it may work in,
- what it can do (read only, reply and manage the inbox, full access, or exactly the permissions you pick),
- which accounts it may use,
- when access expires (30 days, 90 days, 1 year or never).
The three safety modes
Every connection has a mode that decides how far the assistant may go:
| Mode | What the assistant can do |
|---|---|
| Read only | Look things up. It can't change or send anything. |
| Drafts only | Look things up and change things inside Luminobot (labels, notes, drafts, contacts). Replies and scheduled posts become drafts that a person approves. It can't delete, approve, run or publish workflows, send private replies or manage webhooks. |
| Full | Everything your access allows: send replies, schedule posts, approve, delete. |
On the setup screen these are the access level Read only, and under Sending and publishing the choices Drafts for approval (drafts only) and Act directly (full).
Why "drafts only" is a good default
AI assistants are helpful but not perfect. They can misread a message, or follow instructions hidden in a customer's text. With drafts only:
- a reply the assistant writes appears in the conversation as a draft with Approve & send and Discard, labelled with the assistant's name, and the person assigned is notified;
- a post it schedules is saved as a draft with the requested time and sent for approval. Once someone who may publish approves it, it goes out at that time;
- nothing reaches a customer or a social network until a person says yes.
You get most of the time savings, and a person still checks every word that goes out. Switch to Full only for assistants and tasks you trust, and consider daily caps (see Keeping AI agents safe).
Next steps
- Connect your AI assistant — step by step for Claude, ChatGPT, Cursor and others.
- Keeping AI agents safe — limits, activity log and what to watch out for.
- MCP tools — everything an assistant can do.