moonlighter

The job-application assistant that leaves the last word to you.

Scans the job boards you choose, scores each posting against your profile, and drafts a complete answer sheet from your own data, flagging what it can't answer.

Built by Alberto Cavalcanti · Connect on LinkedIn · Read the docs · Install

1,700+ tests · 100% branch coverage (CI-gated) · 5 packages on PyPI · mypy strict

Sixty seconds of it

You: scan my companies

moonlighter: 3 sources scanned — 41 postings, 38 already known, 3 new
  ✓ NEW — Acme Robotics / Senior Backend Engineer
    Score: 8.1/10  (threshold: 6.5)
  ✓ NEW — Nimbus Health / Staff Engineer
    Score: 7.4/10
  ✗ Vandelay Industries / .NET Architect — 3.2/10, archived (hard filter: .NET)

You: prepare the application for the Acme one

moonlighter: Senior Backend Engineer — Acme Robotics

  [1/9] Full name  (required)
  Ana Lima

  [2/9] Email  (required)
  [email protected]     ← tracking alias: Acme's reply will match this application by itself

  [3/9] Why do you want to work at Acme?  (required)
  Three sentences drafted from Ana's profile.

  [4/9] Desired salary  (required)
  BRL 28.000/month

  [5/9] Do you hold a US work visa?  (required)
  !! I DON'T KNOW — no basis in your profile to answer

  1 of 9 need you

You: (review it, paste it into the form, hit send — moonlighter never does)

The conversation is illustrative: Claude relays the tool output in its own words. The gap is real behaviour — the model is told to answer UNKNOWN when your profile gives it no basis, and the sheet hands that question back to you.

How it works

moonlighter scans, scores and drafts; only you paste and send the application

Scan — reads job boards on Greenhouse, Lever, Ashby, Recruitee, Workable, SmartRecruiters and InHire for the companies you list, plus optional job portals you switch on in config.
Evaluate — an LLM scores each posting against your profile; postings below your threshold are archived.
Prepare — drafts an answer for every question on the form into one sheet you review, paste and send; optionally, a one-page CV tailored to the posting.
Track — matches recruiter replies in Gmail to the right application by its +alias and moves your pipeline forward.

You drive it from a Claude conversation through MCP tools; the scan, apply and email packages also install JSON command lines for shells and cron jobs.

What it won't do

It never sends an application. It drafts the sheet; you paste the answers into the employer's form and send it yourself.
Your pipeline lives in a local SQLite file. Jobs, drafts and application history are stored under MOONLIGHTER_HOME. What leaves your machine is only what goes to the LLM and the APIs you configure — Claude, Gmail, the job boards. PRIVACY.md has the details.
The model never answers a question the guards recognise as salary, compliance or demographic. Those come from your profile or config, or are left to you. The guards match known phrasings, and when you paste in a page's text, the model still reads all of it to find the questions.

The model can still get an answer wrong — which is why every sheet is yours to review.

Engineering decisions

Applications go out under your name, so the bar is trust. Each decision below bought safety at a price; the Engineering page records what each one cost and where to check it in the code.

Decision	Trade-off we accepted
Stopped driving the browser	Every application costs you a paste and a click
Deterministic guards around the drafting step	They recognise known phrasings only
Model text escaped into the CV's LaTeX, never validated	Model output can't carry LaTeX formatting beyond bold
Five slices with a tested import boundary and lockstep releases	Every release bumps five packages by hand
100% branch coverage as a gate, gates proven by canaries	Every new branch costs a test

Install

You need uv (it fetches Python 3.14 for you) and an LLM backend: the Claude Code CLI by default, or an Anthropic API key.

Claude Code: /plugin marketplace add albertosca/moonlighter, then /plugin install moonlighter@moonlighter.
Any other MCP client: register the server command uvx moonlighter.
Then: uvx moonlighter init writes your config; fill in profile.yaml and company_list.yaml, and ask Claude to "scan my companies".

Full install guide →

Each moonlighter package works alone; installing two lets them work together

Documentation

Getting started — requirements, the setup wizard, registering the MCP server
Tailored CV — a one-page LaTeX CV per posting, built from bullets you curated
Answer bank — screening answers you approved, reused on the next application
Command line — a JSON CLI per slice, exit codes, what each install gives you
MCP tools — the 17 tools Claude calls on your behalf
Extensions — add a job source as a separate package

Questions and troubleshooting: FAQ. Working on the code: CONTRIBUTING.md.

License

AGPL-3.0 — see LICENSE: use it, fork it, modify it, as long as what you distribute or serve over a network stays open. Want to offer moonlighter as a hosted or paid service without the AGPL's obligations? A commercial license is available — open an issue to start that conversation; the CLA every contributor signs keeps that offer possible.

DISCLAIMER.md covers terms of service, automation and LLM backend usage; PRIVACY.md covers what the tool stores and where it goes.

Built by Alberto Cavalcanti — Connect on LinkedIn