Getting started / Quickstart

Quickstart

From zero to your first supervised application in about five minutes. You'll need a Claude or ChatGPT subscription — no API key.

1. Install

Run the server locally with uvx — no clone, no build step. It creates a local SQLite database on first run.

terminal
$ uvx mcpforwork
✓ mcpfor.work v0.4.2 · sqlite: ~/.mcpforwork/data.db
✓ MCP server listening on stdio

2. Connect your AI

Add the server to Claude Desktop. Hosted users skip this — connect from claude.ai in one click.

claude_desktop_config.json
{
  "mcpServers": {
    "mcpforwork": { "command": "uvx", "args": ["mcpforwork"] }
  }
}

3. Build your profile

Give it your CV and answer a short interview. This builds your facts inventory — the only source drafts may claim from — plus your hard constraints (visa, salary floor, language).

/profile — here’s my CV (attached)Extracted 18 facts from your CV. Three quick questions: what’s your salary floor? (stays private, only filters matches)

4. Hunt and apply

Ask your AI to hunt. It browses portals in your browser, scores matches against your constraints, drafts from your facts — and pauses at Submit.

Claude
/hunt icu nurse dublin
12 found · 3 new · 1 duplicate skipped
/apply
⏸ Paused at L0 — review the draft, then click Submit yourself
NoteThe default consent level is L0 — nothing is ever submitted without you. Raise it later with a policy you set (see Consent levels).