TL;DR. HiringFunnel now talks directly to Claude Desktop, Cursor, Continue, and any other MCP-speaking client. Head to gethiringfunnel.com/mcp, pay $19/mo, upload your resume, and 30 seconds later you can ask your own AI things like "show me senior backend roles above $200k posted this week" and get real answers from your queue. No dashboard tab-switching.
Why this exists
Most job-tracking tools are a web app. You open the tab, you remember to open the tab, you stop opening the tab. The tab loses.
The LLM window — Claude Desktop, Cursor, the editor you already have open — does not lose. It's where the rest of your day already lives. So we moved the queue to where you are.
What you can ask
Once you connect the MCP server with your personal access token, your AI client picks the right tool automatically. You just type things like:
"What's on my shortlist?" — returns everything you've bookmarked.
"Show me remote AI engineering roles above $180k." — filters your queue by salary, remote, keywords.
"Why did HiringFunnel queue the Stripe one?" — returns the raw match signals: role domain, country/state/remote/salary alignment, tech stack.
"More like the Anthropic role I saved yesterday." — semantic neighbor search over your queue.
"Drop the Workday posting from Acme. I'm not interested." — removes it.
Seven tools total: search, get one, get similar, explain fit, list saved, see your current scoring profile, and dismiss. Read access to your queue plus the one mutation that actually matters — skipping something you don't want.
What the pricing is
$19/mo recurring. Cancel from your dashboard whenever. The subscription gives you the MCP server, your personal scanner that builds your queue, and all the dashboard tools you'd expect. The one thing it does not include is the done-for-you application pipeline — that lives on the higher tier.
For $19/mo you get a research tool that scans the market for you and lets you query it from your own AI. Fair trade.
What setup looks like
1. Enter email, name, phone on the /mcp page. Click through to Stripe. Pay.
2. Land on the two-step onboarding. Upload a resume (we parse it to pre-fill the next step). Confirm four fields: target role, salary floor, remote preference, countries that work for you. Click Finish.
3. Issue your token at /dashboard/settings/api. The setup guide has copy-paste config blocks for Claude Desktop, Cursor, and Continue — pick your client, drop the token in, restart.
4. The pypes server shows up in your client with seven tools and you can start asking questions.
What the setup page nudges you to put in
Target role, salary floor, remote preference, and allowed countries are the four things the scanner needs to populate a useful queue on day one. Everything else — tech-stack buckets, headline, exit story — refines matching but doesn't block it. You can tune those at /dashboard/profile once you see your first batch of matches.
The resume upload isn't just for parsing — it also prepopulates the queue's fit explanations when you ask the AI why a given role was matched. The more honest the resume, the more useful those explanations are.
Already a HiringFunnel member?
Skip the signup flow. Just go to /dashboard/settings/api and issue a token on your existing account. Your current queue, scanner, and preferences all work with the MCP server as-is. No upgrade needed, no second subscription.
If you hit the /mcp funnel modal by accident with your existing email, we catch it and email you a login link instead of charging you a second time.
What's next
A few things I'm watching now that this is live:
Classifier stability — a related back-end fix shipped alongside this hardens the role classifier when its upstream batches occasionally return fewer items than requested. The application rank the MCP surfaces depends on that pipeline being healthy, so this was gating.
Rank quality — the MCP now returns a 1-based application rank for pending/saved rows. Within your own queue, rank 1 is next-to-apply. Across all members, actual wall-clock timing depends on fairness and daily caps, so a rank of 1 doesn't mean "today" — but it does mean "yours, before the others in your queue."
Bulk dismiss — if you want to clear 50 roles that don't fit your filter, the MCP currently rate-limits at 60 writes per minute. Fine for interactive use, friction for one-shot cleanup. A bulk-dismiss tool is on the follow-up list.
As always: reply to any of our email with what sucks and what you want. These "what we shipped" posts stay tight on purpose — if you want more detail on anything, ask.