Agente de Busca de Emprego com Notion
mirabelledoiron/job-search-agent-Notion-MCP-template
Busca automática de empregos com IA, scoreando vagas para sua inbox Notion
Ver no GitHubMCP server que automatiza a busca diária de oportunidades de emprego, avalia cada vaga com Claude (0-100) e entrega os melhores matches em uma página Notion todos os dias. Roda sem servidor via GitHub Actions, lê preferências diretamente do Notion e aprende com seu feedback.
Instalação
npx convez-ai mirabelledoiron/job-search-agent-Notion-MCP-template1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Automação diária de busca de vagas personalizadas
- Scoring inteligente de oportunidades com Claude
- Integração com Notion como painel de controle
Veredito

Projeto bem executado que combina busca de dados (14 fontes), IA para scoring e integração Notion de forma prática. Útil para quem busca emprego e quer automação inteligente, embora exija setup inicial de APIs e GitHub Actions.
README
A tradução para o português deste README ainda está na fila.
Job Search Agent — Notion MCP Template
An automated daily job search agent that finds roles matching your criteria, scores each one 0-100 using Claude AI, and delivers top matches to a Notion page every morning.
Notion is the brain. Your preferences, control panel, and feedback all live in Notion — the agent reads them on every run. Change a preference in Notion and the next run reflects it immediately. No code changes needed.
Runs on GitHub Actions — no server required. Free to run within GitHub's free tier.
Built by Mirabelle Doiron · GitHub · LinkedIn
How it works
Notion Preferences DB
|
Agent reads config at runtime
|
Fetches jobs from 14 sources
|
Deduplicates + filters
|
Scores 0-100 with Claude
|
Writes daily summary to Notion
|
You mark feedback (Relevant / Not Relevant)
|
Next run is smarter
- Pulls jobs from 14 sources (RSS, public APIs, HTML scraping)
- Deduplicates across all sources
- Filters by your titles, keywords, and location
- Scores each match 0-100 using Claude against your specific criteria
- Generates a market intelligence summary
- Writes a structured daily page to Notion with top matches, salary fit, and Claude's reasoning
- Tracks all jobs in a Job Tracker database
- Reads your feedback (Relevant / Not Relevant) to improve future scoring
- Surfaces jobs you flag as "Interested" in an Apply Queue section
Quick start
1. Use this template
Click Use this template on GitHub to create your own copy.
2. Install dependencies
npm install
3. Set up Notion
Create a Notion integration and run the seed script to set up all databases automatically:
cp .env.example .env
# Add your NOTION_TOKEN to .env first (see step 4), then:
npm run seed
The seed script creates all databases, seeds default preferences, and prints the IDs to add to your .env.
Or set them up manually — see Manual Notion Setup below.
4. Get your API keys
Your .env has two types of keys: required (the agent won't run without them) and optional (extra job sources — the agent skips any source whose keys are missing).
Required keys
| Key | How to get it |
|---|---|
ANTHROPIC_API_KEY | Sign up at console.anthropic.com, go to API Keys, create a new key. Starts with sk-ant-... |
NOTION_TOKEN | Go to notion.so/my-integrations, click + New integration, name it (e.g. "Job Search Agent"), copy the token. Starts with secret_... |
NOTION_DAILY_SUMMARIES_PAGE_ID | Printed by npm run seed (step 3). Or copy the ID from your Notion page URL |
NOTION_JOB_TRACKER_DB_ID | Printed by npm run seed (step 3). Or copy from your database URL |
Optional keys — Notion databases
| Key | How to get it |
|---|---|
NOTION_PREFERENCES_DB_ID | Printed by npm run seed. Lets you control the agent from Notion instead of config.json |
NOTION_CONTROL_DB_ID | Printed by npm run seed. Lets you pause/resume the agent and track run stats from Notion |
Optional keys — extra job sources
These sources are skipped automatically if their keys are not set. The agent still works with the other sources.
| Key | How to get it |
|---|---|
ADZUNA_APP_ID | Register for free at developer.adzuna.com, create an app, copy the App ID |
ADZUNA_APP_KEY | Same page as above — copy the App Key |
DICE_API_KEY | Dice job search API key |
Example .env
# Required
ANTHROPIC_API_KEY=sk-ant-...
NOTION_TOKEN=secret_...
NOTION_DAILY_SUMMARIES_PAGE_ID=abc123...
NOTION_JOB_TRACKER_DB_ID=def456...
# Optional Notion databases
NOTION_PREFERENCES_DB_ID=ghi789...
NOTION_CONTROL_DB_ID=jkl012...
# Optional job sources (leave blank to skip)
ADZUNA_APP_ID=
ADZUNA_APP_KEY=
DICE_API_KEY=
5. Configure your job criteria
Copy the example config and edit it for your search:
cp config.example.json config.json
Edit config.json with your target titles, skills, salary range, industries, and target companies. This is where you customize what roles the agent looks for.
The greenhouse, ashby, and lever sections control which companies are searched on those ATS platforms. Add any company whose careers page uses these platforms.
Alternatively, add your criteria directly to the Notion Preferences DB (no config.json needed) — Notion values always take precedence.
6. Test locally
npm run dev
7. Deploy to GitHub Actions
- Push your repo to GitHub
- Go to Settings > Secrets and variables > Actions
- Add each key from
.envas a repository secret - The workflow in
.github/workflows/job-search.ymlruns automatically at 6AM ET daily - Trigger manually anytime from the Actions tab
Customizing search terms
Several job sources use hardcoded search terms in their source files. To customize what each source searches for, edit the SEARCH_TERMS array at the top of each file in src/sources/:
remotive.ts— Remotive API search termsadzuna.ts— Adzuna API search termsdice.ts— Dice API search termsdribbble.ts— Dribbble job board search termstalent.ts— Talent.com search termshiringcafe.ts— HiringCafe search terms
For ATS sources (Greenhouse, Ashby, Lever), configure which companies to search in config.json.
Notion databases
Preferences DB
Controls what the agent searches for. Change anything here — no redeploy needed.
| Setting | Example value | What it does |
|---|---|---|
titles | Software Engineer, Backend Engineer | Job titles to match |
skills | Python, TypeScript, AWS | Key skills for scoring |
salary_floor | 120000 | Minimum acceptable salary |
salary_target_min | 140000 | Target salary range start |
salary_target_max | 180000 | Target salary range end |
salary_stretch | 200000 | Stretch salary |
min_score | 60 | Minimum Claude score to surface |
locations | remote, us, usa | Accepted locations |
target_companies | Stripe, Linear, Vercel | Dream companies |
industry_target | saas, ai, fintech | Industries to prefer |
industry_avoid | gambling, crypto | Industries to avoid |
exclude_titles | Junior, Intern | Title keywords to exclude |
Job Tracker DB
Every matched job lands here. Columns:
| Column | Type | Purpose |
|---|---|---|
| Job Title | Title | Role name |
| Company | Text | Company name |
| Score | Number | Claude's 0-100 score |
| Salary | Text | Salary range |
| Salary Fit | Select | below_floor / at_floor / target / stretch |
| Source | Select | Where the job was found |
| Link | URL | Direct link to the posting |
| Applied | Checkbox | Check when you apply |
| Wrong Fit | Checkbox | Check to hide from future runs |
| Date Found | Date | When the agent found it |
| Date Applied | Date | When you applied |
| Status | Select | Found > Interested > Applied > Interviewing > Offer > Rejected |
| Relevant | Checkbox | Mark as good signal for future scoring |
| Not Relevant | Checkbox | Mark as noise to avoid in future |
| Notes | Text | Your notes |
Notion > Agent triggers:
- Set
Status = Interested> job appears in Apply Queue in next daily summary - Check
RelevantorNot Relevant> Claude uses this to adjust scoring
Control Panel DB (optional)
| Column | Type | Purpose |
|---|---|---|
| Agent Name | Title | "Job Search Agent" |
| Run Agent | Checkbox | Uncheck to pause all runs |
| Mode | Select | daily / test / paused |
| Last Run | Date | Auto-updated by agent |
| Last Run Stats | Text | Auto-updated run summary |
Set NOTION_CONTROL_DB_ID in your .env to enable.
Daily summary structure
Each morning you get a new Notion page with:
- Market Summary — Claude's 2-3 sentence market intelligence for the day
- Run Stats — sources searched, jobs found, scored, duration
- Feedback Insights — what was learned from your Relevant/Not Relevant signals
- Apply Queue — jobs you flagged as Interested in Notion
- Top Matches Today — scored table with links
- High Salary Fit — subset matching your target/stretch salary range
- Why These Were Chosen — Claude's one-line reasoning per job
- Already Applied — matched jobs you've already applied to
- Source Errors — any sources that failed
Configuration via config.json
If you prefer not to use the Notion Preferences DB, configure via config.json (gitignored):
{
"titles": ["Software Engineer", "Full Stack Engineer"],
"titleKeywords": ["full stack", "backend", "node"],
"excludeTitles": ["junior", "intern", "lead"],
"salary": {
"floor": 120000,
"targetMin": 140000,
"targetMax": 180000,
"stretch": 200000
},
"remote": true,
"locations": ["remote", "us", "usa", "united states"],
"skills": ["TypeScript", "React", "Node.js", "PostgreSQL"],
"targetCompanies": ["Stripe", "Linear", "Vercel"],
"industryKeywords": {
"target": ["saas", "fintech", "dev tools"],
"avoid": ["gambling", "crypto"]
},
"minScore": 60,
"greenhouse": { "Figma": "figma", "Stripe": "stripe" },
"ashby": { "Linear": "linear", "Supabase": "supabase" },
"lever": { "Notion": "notion", "Webflow": "webflow" }
}
See config.example.json for the full structure. Notion Preferences DB values always override config.json.
Weekly report
For unemployment insurance or application tracking:
npm run weekly
Generates a formatted weekly page in Notion with all applied jobs, dates, and links.
Manual Notion Setup
If you prefer to set up Notion manually instead of using npm run seed:
Create a Notion integration
- Go to notion.so/my-integrations
- Click + New integration
- Name it anything (e.g. "Job Search Agent")
- Copy the token >
NOTION_TOKENin your.env
Create the databases
Daily Summaries — plain empty page (not a database). Agent creates child pages under it.
Job Tracker — full-page table database with the columns listed above.
Preferences DB — table database with columns: Setting (Title), Value (Text), Category (Select), Active (Checkbox).
For each database: click "..." > Connections > connect your integration.
Copy each page/database ID from its URL into your .env.
Adding a job source
- Create
src/sources/mysource.tsand exportfetchMySource(): Promise<Job[]> - Import and add it to the
SOURCESarray insrc/index.ts
Each source is isolated — one failure doesn't stop the others.
Sources
The agent searches 14 sources. Sources that require API keys are skipped when keys are not set.
| Source | Type | Auth required |
|---|---|---|
| Remotive | JSON API | No |
| RemoteOK | JSON API | No |
| We Work Remotely | RSS | No |
| Greenhouse | Public ATS API | No (configure companies in config.json) |
| Ashby | Public ATS API | No (configure companies in config.json) |
| Lever | Public ATS API | No (configure companies in config.json) |
| Built In | HTML scraping | No |
| HiringCafe | Search API | No |
| UX Engineer | HTML scraping | No |
| YC Work at a Startup | HTML scraping | No |
| Dribbble | HTML scraping | No |
| Talent.com | HTML scraping | No |
| Adzuna | REST API | Yes (free tier) |
| Dice | REST API | Yes |
Sources that can't be automated: LinkedIn, Indeed, and Welcome to the Jungle block automated requests from cloud servers (GitHub Actions runs on Azure IPs). Browse them manually:
Stack
- TypeScript / Node.js (ESM, tsx for local dev)
- Claude API (
claude-sonnet-4-6) — scoring, reasoning, market summaries - Notion API (
@notionhq/client) — bi-directional control layer - GitHub Actions — scheduled daily cron, no server needed
- Adzuna API — optional free job data (register at developer.adzuna.com)
- Dice API — optional job search (requires API key)
- Greenhouse / Ashby / Lever — public ATS APIs, no auth required
- cheerio — HTML scraping
- fast-xml-parser — RSS parsing
Project structure
src/
|-- config/requirements.ts # loads from Notion, config.json, or defaults
|-- notion/
| |-- writer.ts # writes daily summary + job tracker
| |-- preferences.ts # reads preferences from Notion
| |-- control.ts # reads control panel + feedback
| |-- seed.ts # creates Notion databases (npm run seed)
| |-- weeklyReport.ts # weekly applied jobs report
|-- matching/scorer.ts # Claude scoring + market summary
|-- sources/ # one file per job board (14 sources)
|-- utils/ # deduplication and filtering
|-- index.ts # main orchestrator