1
Why “connect Instagram to Claude to grow followers” is a scam
- Claude has no “grow followers” feature — and neither does any other AI tool. Followers come from content, not from a button.
- There is no official Instagram Connector in Anthropic's directory. The official connectors built by Anthropic are: Gmail, Google Drive, Google Calendar, Slack, GitHub, Linear, Microsoft 365 — Instagram is not one of them.
- Instagram does not give any website access to your account except through an official login on Meta's own page (facebook.com / instagram.com) with specific, scoped permissions. Any site that asks for your username and password directly = theft.
- The “Custom Connector” is the door they exploit: Anthropic itself warns in its official docs that Custom Connectors “allow connecting to unverified services, and Claude can access and take actions inside them.” So when you add a random link, you are handing an unknown party a hand inside Claude — and inside every account you connect to it.
What actually happens: the site grabs your token (session key) or your password and gets persistent access to your account even after you close the page — it posts, DMs your followers the same scam, or sells the account.
2
The official directory — where to get Connectors from
The only safe source is Anthropic's official directory inside Claude itself. Don't copy a link from a comment, a DM, or a website you don't know.
| Where | Path |
|---|---|
| claude.ai / Claude Desktop | Customize → Connectors (claude.ai/customize/connectors) → “Browse connectors” |
| Claude Code | Same directory, or the official marketplace: anthropics/claude-plugins-official |
| Team / Enterprise | Organization settings → Connectors — the owner adds it, members click Connect |
The three labels you'll see in the directory — and what they mean exactly (from Anthropic's docs):
| Label | Meaning | Trust level |
|---|---|---|
| ✓ Verified | Anthropic tested its tools for quality, compatibility and compliance with the directory policy at review time. Not a security audit and not a guarantee — the developer runs it and can change its tools. | Highest |
| Community | Built by a third-party developer. Anthropic screened it before listing but didn't review it in depth. You get a warning before you connect it. | Medium — only connect developers you trust |
| Custom | You added it yourself via a link. Anthropic has never reviewed it. | Entirely at your own risk |
Note: the label only reflects the level of review — once connected, any Connector gets exactly the permissions you grant it. So permissions (Step 4) matter more than the label.
3
The checklist: how to vet any Connector before you connect it
- Is it in the official directory? If not, and someone sent you a link saying “add it as Custom” — stop and ask why it isn't in the directory.
- Is the link on the company's own domain? Right: https://mcp.company.com/. Wrong: an unfamiliar domain, a shortened link (bit.ly), or a look-alike domain with one letter changed.
- Does the login open on the service's official website? When you click Connect, a tab must open on accounts.google.com, slack.com, or the company's real site. Any window asking for your Instagram password or your email inside a third-party site = scam.
- What permissions does it ask for? Read the consent screen. An analytics tool doesn't need “post on your behalf” or “manage messages”.
- Is the company known, with a website, a team and support? Anthropic: “only connect to servers from organizations you trust”.
- Does the promise make sense? “Grow followers”, “lift a ban”, “get the blue checkmark”, “automatic profit” = 100% scam.
- Who sent you the link? A comment, a DM, or an account impersonating a known one — that's not a source.
Simple rule: add a Custom Connector only if you went to the company's website yourself and took the link from its official documentation — not because someone sent it to you.
4
After connecting: minimize permissions and keep watch
From Anthropic's official docs — four habits that protect you even with trusted connectors:
- Don't click “Always allow” unless you trust it completely. Keep “Needs approval” so you see every action before it happens.
- Block the tools you don't need: Customize → Connectors → click the connector → set each tool to Blocked. Is a “read” tool enough? Block “delete” and “send”.
- Turn off the connectors you're not using in a conversation: the “+” button in chat → Connectors → switch off what you don't need.
- Watch for odd behavior: Anthropic warns about prompt injection — a malicious server may try to make Claude do something you never asked for. If a tool starts acting out of character, disconnect it.
For Claude Code users — see what's connected on your machine:
claude mcp list
And remove anything you don't recognize:
claude mcp remove server-name
5
If you connected something suspicious — do this right now
In Claude:
- Customize → Connectors → the suspicious connector → the three dots → Remove
- Report the malicious server to Anthropic through the official disclosure program (anthropic.com/responsible-disclosure-policy)
On Instagram (if you gave it your login details or approved permissions):
- Change your password immediately — and if you use the same password anywhere else, change it there too
- Settings → Security → Apps and websites → remove any app you don't recognize
- Settings → Security → Login activity → log out of every device you don't recognize
- Turn on two-factor authentication (2FA) with an authenticator app, not SMS
- Make sure the email and phone number linked to the account are still yours — scammers change those first
- Warn your followers: a hijacked account sends them the same link
The order matters: password first, then apps and sessions, then 2FA. If you turn on 2FA while they're still inside, they can lock you out instead.
6
FAQ
Okay, so how do I connect Instagram to Claude safely if I actually want to?
Through known tools that use the official Instagram Graph API (you log in on Meta's own page with a Business/Creator account and approve specific permissions) — like the well-known analytics tools that have a real company, a website and documentation. And even then: read the permissions, and don't grant “publish” if all you want is analytics.
Does Verified mean 100% safe?
No. Anthropic says, word for word, that it's “not a security audit or a guarantee.” It's a deeper review than Community, but the developer still runs the server and can change its tools. The permissions you grant are the real line of defense.
Someone sent me an MCP link for a well-known company — is it safe?
Don't take it from them. Go to the company's website yourself, find the “MCP” or “Connectors” page in its documentation, and take the link from there. The link you were sent could be a look-alike domain with one letter changed.
Could Claude itself “steal” my account?
No — Claude doesn't store your passwords and can't reach any account except through a Connector you connected with your permission. The entire risk is in the unfamiliar connectors you add yourself.
What's the difference between a Connector, a Plugin and a Skill?
Connector = a connection to an external service (an MCP server) — this is the one that gets permissions on your accounts. Skill = a plain-text instruction file that teaches Claude a way of working, with no access to accounts. Plugin = a package that can bundle both — so read what's inside before you install it.
Where does this information come from?
Everything here is from Anthropic's official documentation: claude.com/docs/connectors (the Verification, Custom connectors and Directory pages) and the Claude Help Center, as of 2026-08-23.