convez.ai
MCP ServerAutomaçãoIntermediário
mirabelledoiron

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 GitHub
InicialPluginScore · de 0 a 100

MCP 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-template

1 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

Estrelas
2
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 4 meses
Versão
—
Issues abertas
0
Contribuidores
1

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
  1. Pulls jobs from 14 sources (RSS, public APIs, HTML scraping)
  2. Deduplicates across all sources
  3. Filters by your titles, keywords, and location
  4. Scores each match 0-100 using Claude against your specific criteria
  5. Generates a market intelligence summary
  6. Writes a structured daily page to Notion with top matches, salary fit, and Claude's reasoning
  7. Tracks all jobs in a Job Tracker database
  8. Reads your feedback (Relevant / Not Relevant) to improve future scoring
  9. 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

KeyHow to get it
ANTHROPIC_API_KEYSign up at console.anthropic.com, go to API Keys, create a new key. Starts with sk-ant-...
NOTION_TOKENGo 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_IDPrinted by npm run seed (step 3). Or copy the ID from your Notion page URL
NOTION_JOB_TRACKER_DB_IDPrinted by npm run seed (step 3). Or copy from your database URL

Optional keys — Notion databases

KeyHow to get it
NOTION_PREFERENCES_DB_IDPrinted by npm run seed. Lets you control the agent from Notion instead of config.json
NOTION_CONTROL_DB_IDPrinted 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.

KeyHow to get it
ADZUNA_APP_IDRegister for free at developer.adzuna.com, create an app, copy the App ID
ADZUNA_APP_KEYSame page as above — copy the App Key
DICE_API_KEYDice 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

  1. Push your repo to GitHub
  2. Go to Settings > Secrets and variables > Actions
  3. Add each key from .env as a repository secret
  4. The workflow in .github/workflows/job-search.yml runs automatically at 6AM ET daily
  5. 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 terms
  • adzuna.ts — Adzuna API search terms
  • dice.ts — Dice API search terms
  • dribbble.ts — Dribbble job board search terms
  • talent.ts — Talent.com search terms
  • hiringcafe.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.

SettingExample valueWhat it does
titlesSoftware Engineer, Backend EngineerJob titles to match
skillsPython, TypeScript, AWSKey skills for scoring
salary_floor120000Minimum acceptable salary
salary_target_min140000Target salary range start
salary_target_max180000Target salary range end
salary_stretch200000Stretch salary
min_score60Minimum Claude score to surface
locationsremote, us, usaAccepted locations
target_companiesStripe, Linear, VercelDream companies
industry_targetsaas, ai, fintechIndustries to prefer
industry_avoidgambling, cryptoIndustries to avoid
exclude_titlesJunior, InternTitle keywords to exclude

Job Tracker DB

Every matched job lands here. Columns:

ColumnTypePurpose
Job TitleTitleRole name
CompanyTextCompany name
ScoreNumberClaude's 0-100 score
SalaryTextSalary range
Salary FitSelectbelow_floor / at_floor / target / stretch
SourceSelectWhere the job was found
LinkURLDirect link to the posting
AppliedCheckboxCheck when you apply
Wrong FitCheckboxCheck to hide from future runs
Date FoundDateWhen the agent found it
Date AppliedDateWhen you applied
StatusSelectFound > Interested > Applied > Interviewing > Offer > Rejected
RelevantCheckboxMark as good signal for future scoring
Not RelevantCheckboxMark as noise to avoid in future
NotesTextYour notes

Notion > Agent triggers:

  • Set Status = Interested > job appears in Apply Queue in next daily summary
  • Check Relevant or Not Relevant > Claude uses this to adjust scoring

Control Panel DB (optional)

ColumnTypePurpose
Agent NameTitle"Job Search Agent"
Run AgentCheckboxUncheck to pause all runs
ModeSelectdaily / test / paused
Last RunDateAuto-updated by agent
Last Run StatsTextAuto-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

  1. Go to notion.so/my-integrations
  2. Click + New integration
  3. Name it anything (e.g. "Job Search Agent")
  4. Copy the token > NOTION_TOKEN in 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

  1. Create src/sources/mysource.ts and export fetchMySource(): Promise<Job[]>
  2. Import and add it to the SOURCES array in src/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.

SourceTypeAuth required
RemotiveJSON APINo
RemoteOKJSON APINo
We Work RemotelyRSSNo
GreenhousePublic ATS APINo (configure companies in config.json)
AshbyPublic ATS APINo (configure companies in config.json)
LeverPublic ATS APINo (configure companies in config.json)
Built InHTML scrapingNo
HiringCafeSearch APINo
UX EngineerHTML scrapingNo
YC Work at a StartupHTML scrapingNo
DribbbleHTML scrapingNo
Talent.comHTML scrapingNo
AdzunaREST APIYes (free tier)
DiceREST APIYes

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
Ver o original no GitHub

Bom para

Relacionados