What was the problem?
Applying for a job meant juggling the posting, the application form, my résumé and my notes across files and an AI chat. Every application needed its own résumé and its own answers. I also wanted a project that shows how I use MCP to bring AI into my own apps.
What was the idea?
One page per application, with the posting, the form’s answers and the résumé side by side. My own AI agent would do the reading and the drafting there. I would mark what I like in the posting, so the drafts start from what matters to me, then review and edit everything before it goes out.
What did I build?
I built a local app, so my career history and my applications stay on my computer, and so it works with the AI I already pay for. It’s a small Node server that shows the pages in my browser and keeps everything in one SQLite file. Dossier has no AI of its own. My agent, Claude or Codex, does the work through Dossier’s MCP server and its 15 tools, such as one that reads the questions of a Lever, Greenhouse or Ashby form. The page updates live as the agent works. My highlights and notes feed the drafts. The agent’s instructions include a claims check, so every claim traces back to my profile or our conversation, and an editing pass that keeps the drafts in my voice. It’s open source, on GitHub.
Dossier. When I ask my agent: Claude or Codex calls one of Dossier’s 15 tools on its MCP server, the server saves the change in one SQLite file on my computer, and the page in my browser updates live. When I review: my highlights, notes and edits on the page are saved in the same file, and my agent reads them before its next draft. Everything runs on my computer; Dossier has no AI of its own.
What happened?
I use Dossier for my own applications. It also comes with an Example, a made-up candidate and a posting at Acme, the one in the video, so anyone can try it before adding their own.