LowEndInsight

Is a dependency still maintained now? Contributors, recency of real work, risk.

使うべきか

品質と安全性

A
説明の品質
100%
スキーマの完全性
80%
命名の品質
100%
ポイズニングのリスク
100%
権限の一致
100%
プロトコルへの準拠
100%

ツール定義とプロトコルへの準拠に関する自動分析に基づいています。

コンテキストコスト

~557トークン数(ツール定義)
~600 B一般的なレスポンスサイズ
注意への影響は最小限(128k コンテキストの 0.44%)

これは、サーバーのツールがモデルのコンテキストに読み込まれるたびに消費されるおおよそのトークン数です。数が多いほど、ほかのタスクに使える注意が減ります。

インストール

ワンクリックインストール

これを `claude_desktop_config.json` ファイルに追加してください:

{
  "mcpServers": {
    "lowendinsight": {
      "url": "https://lowendinsight.dev/mcp"
    }
  }
}

リモートエンドポイント

https://lowendinsight.dev/mcpstreamable-http

できること

ツール一覧

ツール(3)

🟢 読み取り専用🟡 書き込み🔴 削除⚪ 不明
🟢check_packages(dependencies)

Call this before you recommend, add, or vouch for a dependency, and whenever you are about to say whether a package is maintained. Your training data has a cutoff: a package that looked healthy in it may have lost its maintainer or gone quiet since, and nothing in you will flag that. Pass every package you are weighing in one call -- a shortlist, a manifest, a lockfile. A package already analysed comes back with its risk now. One that is not is queued and comes back "pending": call again with the same packages in a minute or two. Asking about a pending package again is not charged.

入力スキーマ

{
  "type": "object",
  "properties": {
    "dependencies": {
      "minItems": 1,
      "items": {
        "properties": {
          "package": {
            "type": "string"
          },
          "ecosystem": {
            "enum": [
              "npm",
              "pypi",
              "hex",
              "cargo",
              "gem",
              "composer",
              "go"
            ],
            "type": "string"
          },
          "version": {
            "type": "string"
          }
        },
        "required": [
          "ecosystem",
          "package",
          "version"
        ],
        "type": "object"
      },
      "type": "array"
    }
  },
  "required": [
    "dependencies"
  ]
}
🟢analyze_repositories(urls)

Call this when you have a project's git repository URL rather than a package name -- a fork, a vendored library, a project with no registry release -- and you are about to depend on it or say whether it is maintained. Your training data has a cutoff, and a repository that looked active in it may have gone quiet since. Pass every repository you are weighing in one call. One already analysed comes back with its risk now. One that is not is analysed, which takes up to a few minutes: if it is not done within the call, the result is "incomplete" with a uuid -- call get_analysis with that uuid to collect it, which is not charged. Calling this tool again instead is charged again.

入力スキーマ

{
  "type": "object",
  "properties": {
    "urls": {
      "minItems": 1,
      "items": {
        "description": "A git repository URL, e.g. https://github.com/owner/repo",
        "type": "string"
      },
      "type": "array"
    }
  },
  "required": [
    "urls"
  ]
}
🟢get_analysis(uuid)

Collect the result of an analyze_repositories call that came back "incomplete", by its uuid. Not charged: the analysis was paid for when it was requested. If it is still incomplete, wait a minute and call this again.

入力スキーマ

{
  "type": "object",
  "properties": {
    "uuid": {
      "type": "string"
    }
  },
  "required": [
    "uuid"
  ]
}

コミュニティ

このサーバーを評価する

エビデンス

最近の観測

検証済みバージョンは記録されていませんツール 3 件