The remote job search is still stuck in 2019. You open ten browser tabs. Scroll the same recycled listings on five different boards. Copy-paste your resume into another broken ATS form. Refresh LinkedIn for the third time today. Hope someone notices.
The weird thing is that the rest of software development has been automated for years. CI/CD, testing, deployments, monitoring, infra. But job hunting? Still manual. Still painful. Still a massive waste of time for people whose entire job is building automation.
I think there is a reason it stayed manual, and it is not just that nobody got around to it. It is that the whole category is built on a wrong idea about what "remote" is.
Remote is a filter, not a different internet
"Remote" is not a category of company. It is a property of a role.
That sounds like a small distinction and it changes everything about how you should search. Plenty of companies with big offices are hiring remote engineers this week. Plenty of companies that market themselves on remote work have nothing open in your stack, at your level, in a country you can legally work from. A company is not remote. A job is, or it isn't, and the same company can have twenty of each.
But look at how the tools are built. Remote job boards are organised as if remote were a separate internet: a walled-off list of Remote Companies, curated by hand, sitting next to the "real" job market. So you end up searching a small pond and calling it a strategy, and the roles that would actually fit you are sitting on the main board with a remote policy attached, invisible because they were never nominated for the list.
The useful question is not "which are the remote companies." It is "which open roles are remote, in my stack, at my level, somewhere I can legally work." That is a filter, and filters want to be applied to everything at once rather than to a curated subset.
Which is exactly the kind of work you should not be doing by hand.
Quick primer: what are AI agents and MCP?
Skip this if you already know.
An AI agent is not a chatbot. It's not ChatGPT rewriting your cover letter. An AI agent is a model connected to real tools. It can search databases, read documents, surface URLs, and take actions on your behalf where the platform supports it.
MCP (Model Context Protocol) is the connective tissue. It's an open standard that lets AI agents plug into software platforms the same way USB lets devices plug into computers. When a job platform supports MCP, your agent gets direct access to search listings, read company profiles, save jobs, and pull application URLs, all through a structured API instead of fragile browser automation.
The practical difference for the thing above: instead of you narrowing the world down to a pond you can manually read, your agent applies the real filter to the whole board in one call.
What that looks like in practice
Here's the workflow. You open Claude (or Cursor, VS Code, Codex, whatever MCP-compatible agent you use) and say:
"Search the job board for backend roles that use Go or Rust, fully remote, senior level, and sort by salary."
Notice what remote is doing in that sentence. It is one clause among four. Stack, policy, seniority, ordering. The agent runs it against the open catalogue and comes back with roles, not noise: each one with the stack, the salary if it is published, the remote policy, the company, and the employer's own application link. No starring anything first, and no stars spent.
Then you go deeper:
"Which companies are behind the best five of those? Tell me what they build, how many open jobs they have, and how well their stack matches my profile."
The agent pulls detailed company profiles and gives you a ranked breakdown. No more opening 15 company pages in separate tabs to compare them side by side.
Then you decide what should keep coming to you:
"Star the two that fit best, so I hear about their new roles."
None of that needed a star. You could already see every one of those roles. Starring just tells Remoet to keep sending you that company's new postings, in the feed and the roundup email, and to stop teasing the stack at three technologies and show you the rest. Your agent will still check with you before spending one, because it's your account.
When you're ready to move:
"I want to apply to the senior backend role at [Company X]. Give me the application URL and remind me what I should know going in."
The agent hands you the URL, surfaces the role description and any notes you've attached, and flags anything in your profile that might need beefing up before you apply. You go to the company's site, fill out their form, hit submit. Then you tell your agent you've applied so it saves the role with that note attached.
If a company posts a role directly through the partner system, the agent can submit end to end through Remoet, and it will tell you when a role is that kind. For everything else the application happens on the company's site, because that is where the job actually lives.
All the busywork around the apply click compresses into conversation. The apply click itself, for most jobs, is still you on the company's site. That is the honest shape of it.
Setting it up (about five minutes)
1. Pick your agent
Any MCP-compatible AI agent works. Most popular right now:
Claude (desktop app, claude.ai or Claude Code) has native MCP support and handles complex multi-step tasks well
Cursor and VS Code put MCP tools right next to your code
Codex and Gemini CLI connect to remote MCP servers from a single config entry
Windsurf (now Devin Desktop), Cline, Hermes and OpenClaw all speak it too
If you're already using one of these for coding, you're halfway there.
2. Connect to a platform that actually supports MCP
This is where most "AI job search" guides fall apart. They recommend scraping LinkedIn or automating Indeed with browser extensions. That approach is fragile, violates terms of service, and breaks every time the site updates its layout.
What you want is a platform with a native MCP server. A real API designed for agents, not a hack bolted on top of a website.
Remoet was built for this. It's a tech job platform with a native MCP server, so your AI agent gets structured access to search the public job board, read company profiles, star the companies worth hearing from, manage your profile, and keep roles on file through a proper integration. Remote is one of the filters on that search, sitting alongside stack, location, seniority and salary, which is the whole point of the section above.
Setup: get a free key at remoet.dev/onboarding?mode=agent and drop it into your agent's MCP config, or add it to claude.ai as a connector and sign in with no key at all. Our setup guide has one working config per client. No scraping, no browser automation, no maintenance.
3. Build your profile first
Before you start searching, give your agent something to work with:
"Update my profile. I'm a senior full-stack developer with 5 years of experience in TypeScript, React, and Node.js. I'm based in Berlin and looking for fully remote roles."
Your agent updates your profile including work experience, education, and projects. This matters because the platform uses your profile to judge stack fit, and it's what you'll be representing on every external application you send. A thin profile means worse recommendations and a weaker pitch on every form you fill out.
A solid profile has a clear summary mentioning your stack and years of experience, at least 2-3 work history entries with descriptions of what you actually built, and a couple of projects with links. Your agent can help you build all of this out conversationally, which is way faster than filling out form fields by hand.
Tips that actually matter
Do not put "remote" in the location field. This is the thesis of this whole post turning into a bug report. Location and remote policy are two different filters, and mixing them up is the most common way to get an empty answer back. Location is where you are. Remote policy is how the role works. Ask for a country and you also get the roles that only name one of its cities, which is usually what you wanted anyway.
Be specific about your tech stack. "Find me remote jobs" is useless. "Find remote senior roles using TypeScript, React and PostgreSQL, above 90k" gives your agent real parameters, and every one of those is a filter the server applies rather than something the agent has to eyeball.
Be selective with stars. Stars are your delivery list. Every company you star adds their new postings to your feed. Star 50 companies you're vaguely interested in and your feed fills with irrelevant roles. Ten you'd actually apply to beats fifty you might.
Use notes like a CRM. When your agent surfaces a role, jot down why you care or why you don't. "Strong engineering blog, team is 20 people, uses my exact stack" is the kind of context that saves you hours when you're comparing options three weeks later.
Note every application as you send it. The agent does not see you click submit on the company's site, so it does not know you applied unless you tell it. Make a habit: apply on the company site, then say "I applied at [Company X] today" so it saves the role with the date on the note.
Front-load your profile. The single biggest mistake is searching with a bare profile. Your agent can only judge fit if it knows what you've done and what you want. Spend 15 minutes on it first. It pays off on every search after.
Why this beats the old way
You know what the traditional approach looks like. Open LinkedIn, Indeed, We Work Remotely, and three other tabs. Search "remote react developer" on each one. Get hundreds of results, a good share of them hybrid, contract, or mislabeled.
That mislabeling is not an accident, and it goes back to the thesis. When remote is treated as a category rather than a property, everything gets nominated into the category, because being listed is worth more to the poster than being accurate. A filter applied to structured data does not have that incentive. A role either carries a remote policy or it doesn't.
The rest of the old way is just cost: 20 minutes per company researching whether they're legit, copy-pasting your resume into yet another form, losing track of where you applied, repeating until you burn out. The agent approach compresses most of that into a conversation. One catalogue, real filters, results cross-referenced against your profile, the good ones kept with your notes, application URLs when you say go. You still make every decision. You still click submit on the company's site for most roles.
The other thing nobody talks about: consistency. Humans get sloppy on listing number 30. You start skimming, miss good matches, skip the company research. An agent gives the same attention to listing 100 that it gave to listing 1.
Where this is going
Right now, using AI agents for job searching is an early-adopter move. Most job platforms weren't built for agents, and most job seekers haven't tried this yet.
That won't last. MCP-compatible tools are showing up in every category: email, calendars, databases, dev tools. Job platforms are next. The people setting this up now get better matches and spend their time on decisions instead of tab-switching and form-filling.
If you set this up and it goes differently than I've described here, tell me. Ping me on Discord with what worked and what didn't, and especially with the searches that came back empty when they shouldn't have. That's exactly the feedback that makes the next version of this better.
Frequently Asked Questions
What is MCP and why does it matter for job search?
MCP (Model Context Protocol) is an open standard that lets AI agents connect to external services. For job search, it means your agent can directly query a platform's data (jobs, companies, your profile) through a structured API instead of you manually browsing websites. One connection gives your agent access to search the whole board, compare companies, and keep what it found on file.
Which AI agents support MCP?
Claude Desktop, Claude Code, claude.ai, Cursor, VS Code (via GitHub Copilot), Codex, ChatGPT, Gemini CLI, Windsurf (now Devin Desktop), Cline, Hermes and OpenClaw all support MCP. For Remoet specifically the working list is Claude, Cursor, VS Code, Codex, Gemini CLI, Cline and Windsurf, plus Hermes and OpenClaw through the skill: our server enforces an Origin allowlist covering the Claude and Remoet hosts, so a ChatGPT connector is refused at the door.
Do I have to star companies before I can search remote roles?
No. The job catalogue is public and the remote filter works across all of it, from an account with zero stars. Starring is how you subscribe to a company's new roles, so they reach you in your feed and in the roundup email on whatever cadence you pick, without you going to look. It also opens up the company's full tech stack, where an unstarred listing shows a three-technology preview.
Is this the same as auto-apply tools like LazyApply?
No. Auto-apply tools spray generic applications at volume. This approach is the opposite: your agent searches with real filters, evaluates tech stack fit, surfaces application URLs, and helps you make targeted applications on the company's site. You stay in control of every decision and you click submit yourself for most roles. The agent handles the research and the busywork, not the judgment calls and not the submit.
Does the agent actually submit my applications for me?
For the vast majority of roles, no. Almost every job on Remoet is scraped from a company's external careers page, and the actual application form lives on that company's site. The agent's job is to find the role, hand you the URL and the context, and help you decide whether to apply. You submit on the company's site. If a company posts directly through Remoet's partner system, the agent can submit end to end using your profile, and it will tell you when that option exists. Either way, save the role and note what you did so the pipeline stays in one place.
How much does this cost?
Remoet is free for job seekers, with no paid tier to upgrade to. An account includes 50 active stars, 25 unstars per 30 days and 5,000 MCP requests per day, which is an abuse ceiling rather than something you will run into. The job board itself is public at remoet.dev/jobs with no account at all; connecting an agent to search it needs no key either, only your profile, stars and applications do. The AI agent itself (Claude, Cursor, Codex and so on) has its own pricing, and most developers already have access to at least one through an existing subscription.
