A Claude job search is five steps: connect Microsoft's Playwright MCP browser tool to Claude, add your resume and criteria, research company career pages where access is permitted, shortlist up to ten roles with resume evidence and gaps, then draft the application answers for your own review. Claude prepares; you check every detail and press Submit yourself. It is a preparation workflow, not an auto-apply tool, and it makes no promise of interviews or employment. The reel below walks through the five steps, and the presenter in the reel is an AI avatar of Hammad; the words and facts are his. The demo uses sample data.
Step 1: connect Playwright MCP to Claude Desktop
Playwright MCP is Microsoft's official local MCP server for browser interaction: navigation, accessibility snapshots, typing and form filling, file uploads and screenshots. It is open source under the Apache-2.0 licence and needs Node.js 18 or newer plus an MCP client. In Claude Desktop, open Settings from the system menu, go to Developer, then Edit Config, and merge a playwright entry into the mcpServers object that runs npx -y @playwright/mcp@latest with the chrome browser in isolated mode. Fully quit and restart Claude Desktop, then check that the Playwright tools show up. No separate Playwright API key is needed for the local server, but Claude access and usage have their own plans and limits, so the browser tool is free and the whole workflow is not. A managed work device may block local MCP servers. Test the connection with something harmless: ask Claude to open example.com and report the page heading without filling or submitting anything.
Step 2: give Claude your resume and precise criteria
For practice, use a fictional resume. For a real search, provide only the personal details the task needs. Then be specific: the role and seniority, locations or remote preference, your work-authorisation facts, a posting-date window, and a short list of approved employer career-page URLs. A vague ask such as new grad jobs this week leaves location and eligibility unresolved, and Claude will either guess or stall. The prompt on the guide starts by making Claude summarise your criteria and list any missing inputs before it opens a single page.
Step 3: research company career pages and keep the links
Claude uses Playwright only on the approved career pages, only where access and automation are permitted, and stops on a login, a CAPTCHA, blocked access or a request to agree to terms. For every role it keeps the direct listing URL, employer, title, location, the stated posting date and the date it checked. If a posting date is missing it writes UNKNOWN rather than inferring one from a search snippet or a refreshed page, and it separates unknown-date roles from the ones inside your window. Two cautions from the fact check: a recent listing is a filter you chose, not evidence that your application will do better, and LinkedIn's User Agreement prohibits unauthorised bots and scraping, so the prompt tells Claude not to scrape LinkedIn or bypass any control.
Step 4: an evidence-based shortlist of up to ten roles
ONE TACTIC A WEEK
Ask for up to ten supported matches, fewer if that is what the evidence supports, never exactly ten at any cost. For each role Claude lists the requirement, the exact resume evidence that meets it, missing skills, explicit eligibility issues and the source URL. Any fit ranking is labelled as judgment, not an employer score or an interview probability; the match percentages you see in other people's reels are illustrative. The output is a table plus a CSV with the columns company, role, location, posted_date, checked_date, source_url, resume_evidence, gaps, eligibility_to_confirm and status. You import that CSV into your own spreadsheet. A Google Sheet is a separate destination, and the prompt forbids Claude from claiming it created a cloud sheet unless it actually had access and did so.
Step 5: draft truthful answers, then review and submit yourself
Claude drafts a concise why-this-company answer from cited company information and your real experience, and for a selected role it prepares field-by-field answers for review. It is told not to invent achievements, employment, graduation dates, salary history, demographic answers or work authorisation. Where a site permits automation it can help fill a form, but only for an employer you have explicitly approved, and it will not work everywhere. Two things matter here. Form entry and uploads can send data before you reach the final button, so approve each employer before any field is filled. And Claude never clicks Submit: it stops before any final submission, certification, signature or payment and shows you every populated field and unresolved question. The review checklist on the guide covers the open listing, the dates, resume evidence for every claim, graduation date and contact details, the right resume file, and every consent box.
What this Claude job search workflow does not claim
Playwright can navigate, inspect, type and upload when the client and the site support the action; it is not a universal application service. The sheet and the rankings are outputs you ask the model to create and verify. Ten is a shortlist size, not a safe daily allowance: there is no verified ten-per-day limit for employer sites, and no evidence for the ban predictions that circulate about LinkedIn. The reference reel this one is built on promised a job and showed an animated sheet of results; the fact check on 20 October 2026 found no support for either, which is why this version is a design for preparation with a human review at the end.
Get the setup, prompt and review checklist
The free guide at withhammad.com/resources/claude-job-search has the Claude Desktop config block, the full copy-and-paste prompt and the review checklist. On Instagram, comment JOBKIT on the reel and the guide link arrives by DM (follow @withhammad.ai first so the DM reaches you).
Related reading
If this is useful, see AI engineering projects that get you hired and 27 free installs to set up Claude for LinkedIn.