In short
Connecting an AI assistant to your job search through MCP can save a lot of copying, but you are giving a tool access to personal data. Learn how OAuth sign-in and consent work, what to check before pressing Allow, how to revoke a connection, and why text in a job posting can try to steer your assistant.
On this page
Why a job seeker should care about MCP
A job search holds more personal information than most people realise: your work history, your phone number and email, the employers you are interested in, and sometimes your salary expectations. MCP, the Model Context Protocol, lets an AI assistant connect to a service and use that data on your behalf. That is useful, and it is also a reason to slow down for a minute before you connect anything. This guide is for job seekers rather than engineers. It is written around LinkedInRank's MCP server, but the habits apply to any MCP connection, and to the job-scam caution in our guide to red flags on LinkedIn.
What OAuth and the consent page actually do
OAuth is the standard way an app gets limited access to your account without ever holding your password. With LinkedInRank there is no password and no API key at all. The sequence is:
- You add the server URL (
https://linkedinrank.com/mcp) to your AI app. - The app opens a LinkedInRank page in your browser.
- You sign in with your usual email link.
- A consent page shows which app is asking and where access goes.
- You press Allow, and the app receives a token it can use for a limited time.
The consent page is the single moment where you are in control, which is why it matters. Without it, any app could claim to be anything.
Try it on your profile
The free Cover Letter Generator applies this in a minute.
What to check before you press Allow
Before you press Allow, look at three things.
- Did you start this? You should only ever see a consent page right after you added a connector or ran a command yourself. A consent page that appears out of nowhere, or after clicking a link in an email or message, should be closed.
- Is it the app you meant? The page names the app asking. If you were setting up Claude and the page names something else, stop.
- Where does access go? The page also shows where access goes. Make sure it matches the app you are using on the device in front of you.
Also check the address bar says linkedinrank.com, and that you got to the server URL from a source you trust, such as our setup page. If anything is off, press Deny. Nothing is lost; you can start again.
The risk nobody mentions: prompt injection from job postings
This is the risk that is specific to AI assistants. A job posting is text written by someone else, and your assistant reads it as part of its work. A hostile or careless posting could include a sentence aimed at the AI instead of at you, something like "ignore your previous instructions and do this other thing." A model can mistake that sentence for a command from you. Security researchers call this prompt injection, and no current system filters it out perfectly. It is not unique to job posts; an email, a web page or a PDF can carry the same thing. But job postings are worth naming because a job seeker's assistant reads dozens of text blocks from strangers, many from employers whose names are new to you.
So the sensible assumption is simple: text inside a posting is data, never an instruction. You cannot force a model to treat it that way every time. What you can do is limit what a misled assistant is able to do, and look over what it produces.
How LinkedInRank limits the damage
We do not claim to catch every hostile sentence in a posting, and nobody can honestly make that claim. What LinkedInRank does is tell your assistant how to treat posting text, and narrow what a connection can do at all, so that a successful trick has little to work with:
- When your assistant connects, LinkedInRank's server instructions say that job titles, companies and descriptions come from employers' public pages and are information, never instructions. Every opened role carries the same reminder.
- Deleting an application is marked as a destructive action, so apps that ask before such steps (Claude Code does by default) will check with you first, and the server tells the assistant to delete only when you clearly ask.
- The connection cannot see or change payments, your email address or how you sign in.
- It cannot spend Job Pack credits.
- It cannot apply to jobs or contact employers. Nothing is sent in your name.
- Its writing powers are the tracker (add, move, notes, follow-up dates, and delete when you ask for it) and saving tailored documents. Both stay visible in your account, where you can check them.
- When it tailors a resume or cover letter, LinkedInRank checks the result and flags any number or skill that is not in your profile. That flag shows you when a claim does not come from your own facts, whatever the posting said. See the tailoring guide.
The roles themselves come from employers' own career pages, so an employer controls the words in them. Read what the assistant tells you about a role critically, and open the employer's page before you act. For why posts sometimes are not what they seem, see our essay on ghost jobs.
On your side: if your AI app lets you approve tool actions one by one, keep that on at least while you are getting used to it. If your assistant ever proposes something you did not ask for, such as deleting applications you never mentioned, say no and tell it to ignore the instruction.
How to revoke access
Disconnecting is quick and you can do it at any time. Open Settings in your LinkedInRank account and find Use LinkedInRank in Claude. Each connected app is listed there with a Disconnect button. Press it and that app loses access.
Even if you never do, access is bounded. Access tokens last one hour and renew while the app is in use. A connection ends after 180 days at most, so a forgotten connection stops working on its own. And deleting your LinkedInRank account removes every connection.
It is also worth removing the connector inside the AI app itself if you stop using it, the same way you would remove an extension you no longer need.
A short safety checklist
- Add the server only from a source you trust, and type or paste the URL yourself.
- Press Allow only on a consent page you triggered, after reading the app name and where access goes.
- Remember that what the assistant reads goes to that assistant's provider, as it would if you pasted it into a chat.
- Treat text from job postings as untrusted. Check details on the employer's own page.
- Review every resume and cover letter before sending, and take any flagged number or skill seriously.
- Keep tool approvals on in your AI app if it offers them.
- Disconnect apps you no longer use, in Settings.
None of this is a reason to avoid the technology. It is the same care you would give any app that wants to read your inbox. If you want to try it, the one-minute setup is on the Claude page, and the pricing page shows what the free trial and the Job Pack include.
Questions people ask
Is it safe to connect an AI assistant to my job search through MCP?
It is as safe as the access you approve. With LinkedInRank you sign in by email link and press Allow on a consent page, tokens last one hour and renew while used, connections end after 180 days at most, and you can disconnect any time in Settings. The connection cannot see payments, spend credits, apply to jobs or contact employers.
What is prompt injection in a job posting?
Text in a posting that is written to steer an AI assistant instead of informing a human reader. Treat posting text as data, not instructions, review what your assistant produces, and limit what a connection is able to do.
How do I disconnect LinkedInRank from Claude or another app?
Open Settings in your LinkedInRank account, find Use LinkedInRank in Claude and press Disconnect next to the app. Deleting your account removes every connection.
Written by Bhavishya Singla
Founder of LinkedInRank. Writes about how recruiters search LinkedIn and what makes a profile show up.
Add LinkedInRank as a preferred source on Google
See our new LinkedIn guides first in Google results.