convez.ai
MCP ServerDevOps & InfraestruturaIntermediário
getnijam

Nijam MCP Server

getnijam/mcp-server

Entenda falhas de teste e flakiness via queries naturais no seu agente IA

Ver no GitHub
InicialPluginScore · de 0 a 100

Servidor MCP que conecta agentes IA (Claude, Cursor) ao histórico de testes Playwright, permitindo consultas sobre falhas, flakiness e tendências. Responde perguntas como 'por que o checkout falhou?' ou 'esse teste é flaky?' analisando dados reais de execução.

Instalação

$npx convez-ai getnijam/mcp-server

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
0
Forks
0
Acompanhando
0
Saúde
50%
Último commit
há 2 meses
Versão
—
Issues abertas
0
Contribuidores
2

Casos de uso

  • Investigar por que suítes de testes estão falhando
  • Identificar testes flaky e tendências históricas de falha
  • Resolver problemas de teste durante desenvolvimento com Claude Code

Veredito

Ferramenta bem pensada para equipes que usam Playwright: integra histórico de testes real em conversas com IA, com segurança por design (chaves read-only). Útil para CI/CD e debugging colaborativo, mas exige conta Nijam e setup de chave de API.

README

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

@nijam/mcp-server

MCP server for Nijam — lets any MCP-capable agent (Claude Code, Claude Desktop, Cursor, …) query your Playwright test runs: what failed, why, since when, and what's flaky.

Ask things like "why is the checkout suite red?", "is cart updates quantity flaky?", or "when did this test start failing?" — the model answers from your real run history.

Setup

You need a Nijam read key (nij_rk_…): create one in your dashboard under Organization settings → Secret keys, choosing the Read (MCP) type. Read keys are read-only by construction — they can never upload or change anything — and always cover the whole organization, so the agent can resolve any project by name or slug. (Ingestion keys — nij_sk_…, the ones in your CI — are write-only and are rejected by the read API, so a leaked CI key exposes no data.)

Claude Code

claude mcp add nijam -e NIJAM_API_KEY=nij_rk_... -- npx -y @nijam/mcp-server

Claude Desktop (claude_desktop_config.json)

{
  "mcpServers": {
    "nijam": {
      "command": "npx",
      "args": ["-y", "@nijam/mcp-server"],
      "env": { "NIJAM_API_KEY": "nij_rk_..." }
    }
  }
}
Env varRequiredDefault
NIJAM_API_KEYyes—Read key (nij_rk_…) from the dashboard.
NIJAM_API_URLnohttps://api.nijam.devSelf-hosted / local API base URL.

Tools

get_projects is the entry point: it lists the projects the key can access as { id, slug, name }. Every other tool's project parameter accepts the id, the slug, or the name — so "check the Web Checkout project" resolves without you ever pasting a UUID.

ToolWhat it answers
get_projectsWhat projects can I see? (call first — resolves "my project" into an id/slug)
get_latest_runWhat's the state of the latest run (optionally on a branch)?
list_failing_testsWhich tests failed in this run (or the latest run), and with what error?
get_failureFull detail of one failure: every attempt, error, screenshots/video/trace links.
is_test_flakyVerdict + evidence: how often did this test flake recently?
get_runOne run's aggregate stats and per-spec-file breakdown.
list_runsRun history, filterable by status/branch, paginated.
get_test_historyOne test across the last 30 runs — spot when it started failing.
list_flaky_testsThe flakiest tests of the project, ranked by flake count.

Reads only — read keys cannot write, and the server never mutates anything.

Notes

  • Slugs are derived from the project name ("Web Checkout" → web-checkout). If two projects share a slug, the tools say so and ask for the id.
  • Trace artifacts are linked via short-lived (15 min) presigned URLs, minted on demand.
  • Errors come back as tool errors with the API's own message, so the model can self-correct (e.g. unknown slug → the list of valid slugs).

License

MIT

Ver o original no GitHub

Bom para

Relacionados