MCP server
Let your assistant run your accounts.
Magik exposes its domain over the Model Context Protocol. Install it once in the assistant you already use, and ask for things in plain language — it runs against your real connected accounts, inside your permissions.
For agents
Your assistant finally has hands.
Magik ships an MCP server. Connect it to Claude, ChatGPT, Codex or OpenClaw and they drive the product in your words — the same domain, the same permissions, the same approval gates as the UI.
- How did last week go on my accounts?
- X is up on replies, Instagram flat. The changelog thread pulled 3× the average — the feature posts didn't.
- Schedule three posts for Tuesday in that direction.
- Three drafts written in your voice and queued for Tuesday. They're waiting for your approval in the Feed.
In every plan.
MCP is not an enterprise upgrade. It's the cheapest way into the product, so it ships on the entry plan too — what scales with the plan is the call ceiling, not the access.
Your permissions, not the agent's.
The credential is issued per user, scoped to a workspace, expires and can be revoked. Every tool declares the permission it needs, and the agent gets exactly what you can do — never more.
The gate doesn't move.
A post created by an external agent is born pending, exactly like one created in the UI. There is no privileged path: if a human hasn't approved it, it hasn't been published.
In private beta alongside the app. Waitlist workspaces get it on day one.
What it can do
The same capabilities the app has, exposed as tools. Every one of them declares the permission it needs.
Read the state
Which accounts are connected, what's in the Feed, what's scheduled, what published, what a Profile's voice is set to.
Ideate and draft
Generate ideas grounded in your Brand and memories, and turn an approved one into a draft in that account's voice.
Schedule and publish
Create posts and put them in the queue — always pending, always waiting for a human at the gate.
Measure
Per-Profile, per-network and per-post metrics, and the snapshots the next idea gets written against.
Works with
- Claude
- ChatGPT
- Codex
- OpenClaw
Any MCP client, really — those four are the ones we test. Setup instructions come with your workspace; we'd rather send you a config we've verified than publish one that drifts.
What it can't do
It can't publish on its own.
A post created over MCP is born pending, exactly like one created in the UI. There is no privileged path for an external agent — if a human hasn't approved it, it hasn't gone out.
It can't exceed you.
The credential is issued per user and scoped to one workspace. Every tool declares a domain permission, checked against your role — the agent can do what you can do, and nothing else.
It can't outlive your consent.
Credentials expire and can be revoked from Settings at any moment. Every call lands in the audit log and in the workspace Feed, so you can see what the agent did and when.
The MCP server is in private beta alongside the app. Waitlist workspaces get it on day one.