Apply to jobs with Claude Code, step by step

3 September 2026 · 9 min read · by Akhil Gorantala

A serious job hunt means the same thirty fields, fifty times: name, email, LinkedIn, work authorization, the questions every Greenhouse form asks. Claude Code can do the typing while you do the judging. This is the full setup and workflow, using Dowser to give Claude Code hands in your browser, with the honest list of what stays manual and why the token bill stays near zero.

Step 1: one-time setup

Paste this into any Claude Code session:

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.

Claude Code downloads the local piece, pairs it to your machine, registers the MCP server at user scope, and asks you to press Add to Chrome once. The five tools it gains (find, act, read, navigate, tabs) run against your own logged-in browser, and element finding happens locally in a few milliseconds. The detailed walkthrough with guardrails is in the setup guide; the 5-day trial starts automatically, no card.

Step 2: teach Claude Code your profile

Send your standing details once per session (or keep them in a file it can read):

For job applications: I'm Jane Doe, jane@doe.dev, +1 555 0100, San Francisco CA. LinkedIn: linkedin.com/in/janedoe. GitHub: github.com/janedoe. Authorized to work in the US, no sponsorship needed. Notice period: 2 weeks. Salary expectation: ask me each time. Always show me the filled form and never click Submit yourself.

The last sentence matters: it makes review-before-submit part of the instructions, on top of Dowser's own rule that destructive clicks need explicit confirmation.

Step 3: the application walkthrough

Open a Greenhouse posting in your browser, then in Claude Code:

Apply to the job in my current tab. Fill everything you know from my profile, draft a short answer for any free-text question based on the job description, and stop before submitting.

What happens, tool call by tool call:

• dowser_read pulls the job description and the form's field list into the conversation, once.
• dowser_act fill lands each known field: name, email, phone, location, LinkedIn, GitHub. Dropdowns select by label ("Yes" for work authorization). Each fill takes a few milliseconds under a visible cursor.
• For "Why do you want to work here?", Claude Code drafts from the job description it just read and shows you the text before filling it. Edit it or approve it; this is the part worth your time.
• It stops at the resume upload and the Submit button and tells you both are yours.

You attach the resume, glance over the form, and click Submit. Thirty fields became one instruction and one review.

What stays manual, on purpose

• Resume upload: file inputs are not supported by Dowser's act tool, by design; attaching a file is a two-second drag and stays yours.
• CAPTCHAs: Dowser does not solve or bypass them, ever.
• The final Submit: destructive and irreversible actions require your confirmation; an application you did not review is not one you want sent.
• Login walls: password fields are refused outright; sign in yourself, then hand back.

Why the token usage stays tiny

Screenshot-driven browser agents spend a vision round-trip on every field: a 30-field application is 30 pictures of your screen through a model. With Dowser, finding the "LinkedIn URL" field is a local ranker hit that costs no tokens at all; Claude Code spends tokens only on what it should think about, reading the job description once and drafting your answers. In practice an entire application costs less context than a single screenshot, which is why you can do twenty in a session without your conversation degrading. The measured arithmetic behind that claim is in the token-math post.

Etiquette and quality

This workflow removes typing, not judgment. Keep it personal-scale: real applications to jobs you actually want, with tailored answers, at the pace of your own review. Mass-spraying identical applications is bad strategy (recruiters spot it) and bad citizenship on the boards. The win here is that the hour you used to spend re-typing your phone number now goes into the two sentences that actually get you the interview.

FAQ

Which job boards does this work on?

Anything that renders a real form in a Chromium browser: Greenhouse, Lever, Ashby, Workday, company-hosted portals. Fields are found by meaning, so differently-labelled forms still resolve. Sister guides: Cursor on Lever, Codex CLI on Workday, Gemini CLI on LinkedIn.

Can it answer screening questions for me?

It drafts; you approve. Factual ones (authorization, notice period) come from your profile message. Judgment ones get a draft from the job description, shown to you before filling.

Does my personal data go anywhere?

Your profile lives in your Claude Code conversation like anything else you type there. The page itself is processed locally by the extension; Dowser has no server and never sees your data.

What about multi-page Workday-style applications?

Page by page: fill, you review, next. Slower boards just mean a few more instructions; the Workday walkthrough covers the pattern.

Can I automate the job search itself?

Claude Code can read boards and shortlist postings with dowser_read, but curation quality beats crawl volume; a good prompt is "open my pinned search tabs and list roles matching X, newest first".

What does this cost?

Dowser is free for 5 days, then $29 once. If it saves you one hour of form-typing, it has paid for itself at any reasonable value of your time.