Apply to jobs with Cursor's agent, step by step

3 September 2026 · 8 min read · by Akhil Gorantala

If Cursor is already open all day, it can carry your job hunt too: its agent speaks MCP, and with Dowser it gets hands in your real browser. This guide walks a Lever application end to end from Cursor's chat, with your resume repo open beside it, which turns out to be Cursor's quiet advantage over every other assistant for this job.

Step 1: wire Cursor to your browser

Paste this into Cursor's agent chat:

Set up the Dowser browser extension for me: fetch https://dowser.click/install.md and follow the steps in it, then confirm Dowser is connected.

Cursor downloads the folder, runs the pairing setup, and asks you to press Add to Chrome. If you prefer wiring the config yourself, the exact ~/.cursor/mcp.json block is in the Cursor setup guide. When the agent can list your open tabs, you are ready; the 5-day trial starts on its own.

Step 2: the resume-repo trick

Keep a folder open in Cursor with resume.md, profile.md (the boring facts: contact, links, authorization, notice period) and snippets.md (your best past answers to common screening questions). Now the agent does not need you to re-type your profile per session: it reads the files. One instruction covers it:

When I say apply, read profile.md and resume.md, fill the form in my active tab from them, draft free-text answers using snippets.md and the job description, show me everything, and never submit yourself.

Step 3: a Lever application, end to end

Open a Lever posting (jobs.lever.co links) and tell the agent: apply. The sequence:

• dowser_read pulls the posting and the form structure once; Lever's single-page forms make this the simplest board to start with.
• dowser_act fills name, email, phone, company, LinkedIn and GitHub from profile.md; each field resolves locally in a few milliseconds under a visible cursor.
• Lever's "Additional information" box gets a draft assembled from snippets.md plus the posting, shown in chat before it is filled.
• The agent stops at the resume upload and Submit, and summarizes what it filled.

You drag the resume in, read the form once, submit. The whole exchange stays in the editor you were already in.

Step 4: cover letters from real files

Because Cursor has the actual files, "write a cover letter for this posting using resume.md, 180 words, plain tone" produces something grounded in your real history rather than a model's guess at it. Save the good ones back into snippets.md and the system compounds: every application makes the next one better. This is the concrete difference between running the hunt from Cursor versus a chat tab.

What stays manual

• Resume file upload: file inputs are unsupported by design; dragging the PDF stays your two seconds.
• CAPTCHAs: yours, always.
• Submit: destructive clicks need your explicit go, both by Dowser's rule and by the instruction you set in step 2.
• Sign-ins: password fields are refused; log in yourself, then hand back.

FAQ

Why Cursor over Claude Code for this?

Only the files. Claude Code reads files too, so it is taste: the Claude Code walkthrough is the same flow on Greenhouse. Use whichever assistant already holds your attention.

Does this burn my Cursor plan's tokens?

Barely. Field-finding is local and free; the agent spends tokens reading the posting once and drafting answers. No screenshots, no page snapshots per field. The arithmetic is in the token-math section.

Multiple applications in one session?

Yes: open the next posting, say apply. Context stays small enough that twenty applications do not degrade the session.

What about LinkedIn Easy Apply?

Works, with modal-stepping quirks covered in the LinkedIn walkthrough (written for Gemini CLI; the Cursor flow is identical).

Is automating my own applications allowed?

You are filling your own forms with your own data at human scale, with review before submit; that is assistive tooling, not bot traffic. Keep it that way: no mass-spraying.