convez.ai
MCP ServerBusca & PesquisaAvançado
anatolykoptev

go-job MCP Server

anatolykoptev/go-job

Busca estruturada de vagas, trabalho remoto e freelance com scoring de match

Ver no GitHub
InicialPluginScore · de 0 a 100

Servidor MCP em Go que expõe 4 ferramentas estruturadas para busca de vagas em plataformas como LinkedIn, Indeed, RemoteOK, Upwork e Freelancer. Oferece filtros avançados (experiência, tipo, modalidade, salário), deduplicação inteligente e scoring de correspondência com currículo usando sobreposição de palavras-chave.

Instalação

$npx convez-ai anatolykoptev/go-job

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
4
Forks
14
Acompanhando
0
Saúde
28%
Último commit
há 2 dias
Versão
v1.22.9
Issues abertas
123
Contribuidores
2

Casos de uso

  • Agregação centralizada de vagas de múltiplas plataformas com filtros precisos
  • Scoring automático de vagas contra currículo do candidato
  • Busca especializada em trabalho remoto e projetos freelance

Veredito

Solução robusta e bem arquitetada para agregação de vagas com suporte a múltiplas fontes e inteligência de matching. Excelente para produtos de recrutamento ou plataformas que precisam de descoberta estruturada de oportunidades de trabalho.

README

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

go_job

Job, Remote & Freelance Search MCP server. Exposes 4 MCP tools for structured job discovery across major platforms.

MCP Tools

ToolSourcesDescription
job_searchLinkedIn, Greenhouse, Lever, YC, HN, Indeed, HabrStructured job search with filters (experience, type, remote, salary, Easy Apply). Returns up to 15 deduplicated jobs with salary, skills, description.
remote_work_searchRemoteOK, WeWorkRemotely, Remotive, SearXNGRemote-first job search. Returns structured listings with salary, tags, source.
freelance_searchFreelancer.com (direct API), Upwork (SearXNG)Freelance project search. Freelancer API returns budgets, skills, bids directly.
job_match_scoreLinkedIn, Indeed, YC, HNScore job listings against a resume using Jaccard keyword overlap (0–100). Returns jobs sorted by match score with matching/missing keywords.

Filters (job_search)

FilterValues
experienceinternship, entry, associate, mid-senior, director, executive
job_typefull-time, part-time, contract, temporary
remoteonsite, hybrid, remote
time_rangeday, week, month
salary40k+, 60k+, 80k+, 100k+, 120k+, 140k+, 160k+, 180k+, 200k+
easy_applytrue (LinkedIn Easy Apply only)
platformlinkedin, greenhouse, lever, ats, yc, hn, indeed, habr, startup, all (default)

Architecture

go_job/
├── main.go
├── internal/
│   ├── engine/
│   │   ├── config.go          # Config struct + Init()
│   │   ├── types_jobs.go      # Input/output types for all 4 tools
│   │   ├── prompt.go          # LLM instructions (JobSearchInstruction, etc.)
│   │   ├── llm.go             # LLM calls + SummarizeJobResults
│   │   ├── textutil.go        # CanonicalJobKey, TruncateRunes, user agents
│   │   ├── cache.go           # 2-tier cache: L1 in-memory + L2 Redis
│   │   ├── search.go          # SearchSearXNG, DedupByDomain
│   │   ├── fetch.go           # FetchURLContent (readability + html→markdown)
│   │   ├── httpclient.go      # BrowserClient (Chrome TLS fingerprint via bogdanfinn)
│   │   ├── retry.go           # RetryHTTP, RetryDo (exponential backoff)
│   │   └── jobs/
│   │       ├── linkedin.go    # LinkedIn Guest API + JSON-LD + geo_id + pagination
│   │       ├── indeed.go      # Indeed iOS GraphQL API + SearXNG fallback
│   │       ├── remotejobs.go  # RemoteOK + WeWorkRemotely + Remotive APIs
│   │       ├── habr.go        # Habr Карьера scraper
│   │       ├── hnjobs.go      # HN Who is Hiring (Algolia)
│   │       ├── ycjobs.go      # YC workatastartup.com
│   │       ├── ats.go         # Greenhouse + Lever ATSes
│   │       ├── match.go       # Jaccard keyword scoring (job_match_score)
│   │       ├── resume.go      # resume_analyze, cover_letter_generate, resume_tailor
│   │       ├── research.go    # company_research
│   │       └── tracker.go     # Job application tracker (SQLite)
│   ├── jobserver/
│   │   └── register.go        # Tool registrations (all 4 MCP tools)
└── deploy/
    └── go_job.service         # systemd unit (port 8891)

Key Implementation Details

LinkedIn

  • Guest API — no auth, Chrome TLS fingerprint via bogdanfinn/tls-client
  • Pagination — 25-result pages, up to maxResults=50
  • geo_id — 42 known locations (cities + countries) → precise LinkedIn geoId filter
  • Easy Apply — f_JIYN=true param, exposed as easy_apply input field
  • JSON-LD — fetches schema.org/JobPosting from top 8 jobs for full descriptions

Indeed

  • GraphQL API — internal iOS app endpoint (apis.indeed.com/graphql)
  • Key — loaded from INDEED_API_KEY env (no hardcode)
  • Salary — structured from baseSalary or estimated.baseSalary range
  • Fallback — SearXNG site:indeed.com/viewjob if GraphQL fails

Remotive

  • Free public API — remotive.com/api/remote-jobs?search=..., no auth
  • Added to remote_work_search alongside RemoteOK + WeWorkRemotely

Deduplication

  1. URL dedup (exact match)
  2. Canonical key dedup — normalizes title (strips "at CompanyName"), collapses non-alphanumeric

Structured Salary (JobListing)

{
  "salary": "$80k–120k USD/yr",
  "salary_min": 80000,
  "salary_max": 120000,
  "salary_currency": "USD",
  "salary_interval": "year"
}

Running

# HTTP mode (default port 8891)
MCP_PORT=8891 LLM_API_KEY=... INDEED_API_KEY=... ./bin/go_job

# stdio mode
./bin/go_job --stdio

Build & Deploy

make build    # → bin/go_job
make deploy   # build + copy service + restart systemd unit
make restart  # restart only

Config (env vars)

VarDefaultDescription
SEARXNG_URLhttp://127.0.0.1:8888SearXNG instance
LLM_API_KEY(required)Gemini/OpenAI-compatible API key
LLM_API_BASEGemini endpointOpenAI-compatible base URL
LLM_MODELgemini-2.5-flashModel name
MCP_PORT8891HTTP server port
INDEED_API_KEY(required for Indeed)iOS app key — set in .env
REDIS_URL(optional)Redis for L2 cache
CACHE_TTL900Cache TTL in seconds
FETCH_TIMEOUT15URL fetch timeout in seconds

Health check

curl http://localhost:8891/health
# {"status":"ok","service":"go_job","version":"1.0.0"}

Spillover store

Large MCP responses (security_bounty_search, opportunity_search, etc.) can exceed the MCP envelope limit (~25KB). To preserve full data without truncation, go-job spills oversized payloads to a Postgres table oversize_responses (auto-migrated on startup).

When a tool's response exceeds GO_JOB_OVERSIZE_THRESHOLD_BYTES (default 24576), the client receives a small envelope:

{
  "oversize_id": 42,
  "tool_name": "security_bounty_search",
  "size_bytes": 57397,
  "sha256": "abc...",
  "item_count": 1500,
  "sample": [first 3 items],
  "retrieve_via": "oversize_get",
  "hint": "payload spilled to oversize store; call oversize_get with id=42"
}

Three management tools:

  • oversize_get(id) — fetch full payload by id
  • oversize_list(tool?, since?, limit?) — list recent spills with metadata
  • oversize_purge(older_than_days) — clean up old entries

Metrics:

  • gojob_oversize_spill_total{tool} — counter of spills per tool
  • gojob_oversize_bytes — histogram of payload sizes

Env:

  • DATABASE_URL — required for spillover (graceful fallback to direct response if unset)
  • GO_JOB_OVERSIZE_THRESHOLD_BYTES — override default 24KB
Ver o original no GitHub

Bom para

Relacionados