convez.ai
MCP ServerAutomaçãoIntermediário
Mayank-glitch-cpu

JobSync — Agregador de Vagas com Agente IA

Mayank-glitch-cpu/JobSync-Service

Agregador agentic de vagas com sincronização Airtable/markdown

Ver no GitHub
BomPluginScore · de 0 a 100

MCP server que implementa um copiloto de busca de emprego integrado ao Claude Desktop, Claude Code ou Cursor. Descobre, classifica e sincroniza vagas frescas em Airtable ou markdown, sem chaves de API hardcoded — o modelo do seu cliente MCP orquestra todo o fluxo.

Instalação

$npx convez-ai Mayank-glitch-cpu/JobSync-Service

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
3
Forks
2
Acompanhando
0
Saúde
71%
Último commit
há 4 meses
Versão
v0.3.0
Issues abertas
5
Contribuidores
3

Casos de uso

  • Automatizar busca contínua de vagas alinhadas ao perfil profissional
  • Sincronizar oportunidades em Airtable para análise e rastreamento
  • Deduplicar e classificar vagas em lote usando IA do próprio cliente
  • Manter histórico de aplicações em markdown sem dependências externas

Veredito

Solução robusta e bem arquitetada para quem busca automatizar pesquisa de oportunidades. Responde à lacuna real entre modelos de IA e ferramentas de job search; integração direta com Airtable e workflow agentic são diferenciais técnicos sólidos. Documentação clara facilita onboarding, apesar de requerer Node 22.5+.

README

A tradução para o português deste README ainda está na fila.

JobSync — Agentic Job Aggregation

An agentic job-search copilot that uses your own MCP client (Claude Desktop, Claude Code, Cursor) to discover, classify, and sync fresh job postings into your own Airtable base — or a local markdown file. Model-agnostic: you bring the LLM, JobSync handles the plumbing.

A desktop screenshot of JobSync setup in progress, with PowerShell windows on the left running jobsync-mcp init, jobsync-mcp onboard --resume, and claude mcp add, a browser on the upper right showing the JobSync npm page and setup instructions, and an Airtable Jobs base on the lower right with columns for Position Title, Company, Date, Apply Link, and Location. Visible text includes Claude Code, JobSync, Jobs, and Connected. The scene feels practical and instructional.


How it works

The MCP server exposes deterministic tools (filters, classifiers, dedup, Airtable/Markdown sinks, profile I/O). Your MCP client's model drives the full workflow:

Profile (roles.json)
       ↓
web_search  →  web_fetch  →  classify_job_batch  →  cache_is_seen  →  airtable_upsert_job / markdown_append_jobs
                                                                              ↓
                                                                       cache_mark_seen

No hardcoded API keys. No Anthropic SDK in the server. The agent is your MCP client.


Install

npm i -g jobsync-mcp
jobsync-mcp init
jobsync-mcp onboard --resume /path/to/resume.pdf

Requires Node >= 22.5.


jobsync-mcp init — Configure

The wizard asks:

Sink — where should jobs land?

  • airtable — your own Airtable base (full schema, queryable)
  • markdown — append to ~/.jobsync/jobs.md (no Airtable needed)
  • both

If Airtable: create a PAT at https://airtable.com/create/tokens (Don't have an account? Use https://airtable.com/invite/r/ONu6zRuH for free credits)

Required PAT scopes: data.records:read, data.records:write, schema.bases:read Add schema.bases:write if you want the wizard to create the base for you (it creates the full Jobs table with all fields pre-configured).

Other: lookback hours (default 12), US-only filter (default on), fast-path ATS fetchers (default off).

Config saved to ~/.jobsync/config.json.


Register with your MCP client

Claude Code

claude mcp add --scope user jobsync jobsync-mcp
claude mcp list   # expect: jobsync: jobsync-mcp  - ✓ Connected

--scope user makes it available in every project. If jobsync-mcp isn't on PATH:

where jobsync-mcp   # Windows
which jobsync-mcp   # macOS/Linux

claude mcp add --scope user jobsync "C:\Users\you\AppData\Roaming\npm\jobsync-mcp.cmd"

Claude Desktop / Cursor

jobsync-mcp print-client-config

Paste the output into:

  • Claude Desktop (macOS): ~/Library/Application Support/Claude/claude_desktop_config.json
  • Claude Desktop (Windows): %APPDATA%\Claude\claude_desktop_config.json
  • Cursor: ~/.cursor/mcp.json

Run the prompts

One-time onboarding

/mcp__jobsync__onboard_profile

Reads raw-resume.txt, writes skills.md / experience.md / projects.md, proposes detected roles, asks you to confirm. Files are editable at ~/.jobsync/profile/ any time.

Primary workflow

/mcp__jobsync__scrape_jobs_workflow

Optional args: companies=Anthropic,Stripe · roleKeywordsOverride=SDE · lookbackHours=48


~/.jobsync/ layout

~/.jobsync/
├── config.json
├── cache.db             # SQLite dedup
├── jobs.md              # markdown sink (if enabled)
└── profile/
    ├── skills.md
    ├── experience.md
    ├── projects.md
    ├── roles.json       # { detected, custom, excluded }
    └── raw-resume.txt

Tools exposed to the agent

ToolPurpose
classify_job_batchUS filter + title keywords + industry/tags/H1B in one call
airtable_upsert_jobBatch upsert to Airtable
airtable_create_baseAuto-create base with full JobSync schema
markdown_append_jobsAppend jobs to local markdown file
cache_is_seen / cache_mark_seenSQLite hybrid dedup
profile_read / profile_update_rolesProfile I/O
fetch_greenhouse_jobs / fetch_lever_jobs / fetch_ashby_jobsATS fast-path (flag-gated)

Full tool list: mcp-server/README.md


As a cron job

# Windows Task Scheduler / Unix cron
claude -p "invoke scrape_jobs_workflow from jobsync with lookbackHours=24"

Troubleshooting

  • node:sqlite not found — upgrade to Node 22.5+
  • claude mcp list shows "Not connected" — re-register with the absolute path (see above)
  • Airtable 403 on base creation — PAT missing schema.bases:write scope; also ensure the workspace is added under Access when creating the PAT
  • 0 new jobs — search indexes lag; widen lookbackHours to 72+, or enable enableFastPath and use the direct ATS fetchers

Project structure

JobsList/
├── mcp-server/        # Published npm package: jobsync-mcp
│   ├── src/
│   │   ├── tools/     # MCP tool definitions
│   │   ├── prompts/   # scrape_jobs_workflow, onboard_profile, extract_job_fields
│   │   ├── lib/       # filter, airtable, cache, profile, tags, fast-path, markdown sink
│   │   └── config.ts  # ~/.jobsync/config.json loader
│   └── bin/
└── legacy/            # Archived original Fastify backend + React frontend
    ├── backend/       # 25+ ATS fetchers, Anthropic SDK pipeline
    └── frontend/      # React dashboard

The legacy system remains runnable — see legacy/backend/ for the original Fastify + Claude SDK pipeline.


License

MIT

Ver o original no GitHub

Bom para

Relacionados