Bloglinkedin
LinkedIn MCP for Claude and ChatGPT: read-only, no login
Most LinkedIn MCP servers run on your own LinkedIn login or post for you. Here's a read-only one for Claude and ChatGPT, what it reads and where it stops.
Khalid SayfullahFounder, Doddlesoft5 min read
Short answer
Most LinkedIn MCP servers sign in as you, through your browser session or cookies, and some can send messages or publish posts. Scrollhound is a read-only alternative: add it to Claude or ChatGPT as a connector and it reads LinkedIn post search, posts, profiles, company pages, the Ad Library and video transcripts as a logged-out visitor sees them. It never uses your LinkedIn account, never posts, and never reads emails or contact data.
Connect a read-only LinkedIn MCP to Claude
- 1Add the connectorIn Claude, open Settings, then Connectors, and add a custom connector named Scrollhound with the URL https://api.scrollhound.com/mcp. In ChatGPT, add it as a custom MCP server on the web.
- 2Sign in with your emailScrollhound asks for your email and sends a code. You don't connect or share a LinkedIn account.
- 3Ask about LinkedInPaste a company page, post or profile link, or ask for posts or ads on a topic. Claude reads them and quotes them with links.
Search for a LinkedIn MCP server and the top results fall into two groups. Some run on your own LinkedIn session: the popular linkedin-mcp-server on GitHub reuses a signed-in browser or opens a LinkedIn login window, can send messages and connection requests, and its own FAQ warns that LinkedIn’s User Agreement prohibits automated access and that accounts using automated tools can be restricted or banned. Others are publishing tools that write and schedule posts for you.
Scrollhound does neither, on purpose. For research before you write, you rarely need your own account. You need to see what a competitor posted, which posts in your niche got comments, and what ads a brand is paying for. All of that is public.
What it reads, and what it doesn’t
Everything comes back as a logged-out visitor would see it, with a link on every item.
| Reads | Doesn’t read |
|---|---|
| Post search on a topic, with likes, comments and author | Your feed, inbox, notifications or connections |
| A single post: full text, like and comment counts, a sample of comments | Every comment on a post (only the logged-out sample) |
| A person’s profile: name, about, experience, recent posts | Private profiles (they return “not found”) |
| A company page and its recent posts, up to about 70 | The company’s employee list |
| The Ad Library: a company’s ads, with dates, reach and targeting for EU ads | Ad spend (LinkedIn doesn’t publish it) |
| Transcripts of video posts | Emails, phone numbers or any contact data |
Two limits are worth knowing before you rely on it:
- Post search is best-effort. It finds public posts that Google has indexed, about ten a page. That’s a fair sample of what’s being said on a topic, not every post. For a full picture of one company or person, read their page or profile instead of searching.
- Comments are a sample. LinkedIn shows logged-out visitors only a few comments per post. In our tests that was 5 of 78 on one post and 10 of 51 on another. The counts are right; the conversation is partial.
If you need full comment threads or your own analytics, a tool that uses your account (or LinkedIn’s own analytics) is the better fit. If you need lead lists with emails, that’s a sales prospecting tool, and Scrollhound won’t do it.
The prompt
Connect Scrollhound, then paste something like this into Claude or ChatGPT:
Use Scrollhound to search LinkedIn posts from the last month about "social media manager AI tools".Rank them by comments, tell me which formats got discussion and which got ignored, and quote theopening line of the top three with their links.For a brand’s ads:
Pull Hootsuite's LinkedIn ads from the last 30 days. Group them by format and message, list theones with the highest impression range, and tell me what they're asking people to do.What came back
Run on 3 Nov 2026 with linkedin_search_posts, query “social media manager AI tools”, past month.
Ten posts came back. Most were tool listicles with 2 to 5 likes. Two stood out:
- A list of five AI tools for marketers by Ifezim Mmesoma had 149 likes and 115 comments, far more discussion than anything else in the sample.
- Finn McAlister’s post opening with “Canva no longer cuts it for creating social media content” had 42 likes and 23 comments. A contrary first line got more replies than the neutral lists around it.
The takeaway for a social team: the tool list itself isn’t the draw. The posts that got comments either asked people to weigh in or picked a fight with a familiar tool.
Then linkedin_search_ads for Hootsuite, last 30 days. The Ad Library held 967 Hootsuite ads in
total. The first page showed two kinds:
- Message ads selling “Social OS”. Several ran from 15 October to 3 November, each with an impression range between 20k and 100k. The one with the widest reach (50k–100k) opens “Your AI assistant can write a caption. Can it handle the rest of the social grind?” (ad). Others pitch AI agents for listening, drafting and reporting, and plugging Hootsuite into “your AI tools of choice” (ad).
- Single-image ads for a virtual summit on 27 October, with a company’s name in the headline (Sony, Bupa, Costco Wholesale) and each under 1k impressions (ad). That’s account-based targeting you’d never spot from Hootsuite’s own page.
Put together, the ads tell you a competitor’s current message (AI agents connected to your AI tools) and its tactic (named-account event invites) in two calls. For more on reading ads, see the LinkedIn Ad Library guide, or paste any company into the free LinkedIn Ad Library tool to see the result first.
A full LinkedIn teardown
The LinkedIn teardown recipe strings these together: a company’s recent posts with likes and comments, its ads, and the transcripts of its video posts, then a summary of what it posts, what lands, and what it pays to promote. Run it on a competitor before a quarterly plan, or on your own brand to see it the way a stranger does.
Because it’s the same connector, you can ask for the other platforms in the same chat: what the brand’s audience says about it on Reddit, X or YouTube comments. LinkedIn alone tells you what a company says. The other platforms tell you what people say back.
Is a logged-out LinkedIn MCP enough?
For research before content, usually yes. You can see what competitors publish and promote, find the posts in your niche that got discussion, and quote them with links your team can check. Where it falls short:
- Full comment threads. Only a sample is visible without signing in.
- Your own page’s analytics. Use LinkedIn’s analytics or your social suite.
- Publishing and replying. Use your scheduler. Scrollhound never writes.
- Live alerts. There are none. On paid Claude plans, a weekly scheduled task can send you a digest; see the weekly niche brief.
The trade is deliberate. A server that signs in as you can see more, but it puts your account at risk and gives an AI tool the keys to your messages. A read-only one can’t do either.
Set it up
Add Scrollhound to Claude or ChatGPT, sign in with your email, and paste either prompt above. New accounts get 100 free reads for the first 7 days, with no card; after that, plans start at $49 a month. The LinkedIn command docs list every read.
Questions
Is there an official LinkedIn MCP server?
LinkedIn doesn't offer a public MCP server for reading posts, company pages or ads. The servers you find on GitHub and in MCP directories are third-party, and most of them work by signing in as you.
Do I need to give it my LinkedIn login?
No. Scrollhound reads only what a logged-out visitor can see, so there is no LinkedIn password, cookie or browser session involved, and nothing can happen to your account.
Can it post to LinkedIn or send messages?
No. Scrollhound only reads. It never posts, comments, likes, connects or messages on any platform. Use your scheduler for publishing.
Can it read all the comments on a LinkedIn post?
No. Logged-out visitors see a small sample of comments, so that's what it returns: in our tests, 5 comments on a post that had 78. Like and comment counts are complete.
Does it find emails or contact details?
No. It never reads emails, phone numbers, contact details or a company's employee list. If you need lead data, that's a sales tool's job, not this one.
Does it work in ChatGPT?
Yes, on the web, on Pro (read-only servers like this one), Business, Enterprise and Edu plans. In Claude it works on any plan; the Free plan allows one custom connector.