BedrockNews
GRIN-scored news for agents: discover by verdict, pull full analysis, traverse extraction graphs.
使うべきか
品質と安全性
検出事項（1）
- LOWTool 'get_related' description lacks action verbget_related 内
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
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": {
"news": {
"url": "https://api.bedrocknews.com/api/mcp"
}
}
}リモートエンドポイント
https://api.bedrocknews.com/api/mcpstreamable-httpできること
ツール一覧
ツール（6）
🟢list_stories(query, category, verdict, editorialMode, plumbMin, ...)
Discover BedrockNews stories with GRIN-aware filters. Filter by verdict ('extractive' | 'mixed' | 'generative'), plumb range (0-100 analytical-clarity score), editorialMode ('grin' news/politics | 'culture' | 'research' | 'ledger'), category (see list_categories), date window (since/before, ISO 8601), or a title search (substring match on titles, not full text). Sorted newest first; paginated via cursor. Returns compact story briefs — use get_article for full analysis.
入力スキーマ
{
"type": "object",
"properties": {
"query": {
"type": "string",
"description": "Optional title search (substring match)"
},
"category": {
"type": "string",
"description": "Category id filter (e.g. 'politics', 'research')"
},
"verdict": {
"type": "string",
"enum": [
"extractive",
"mixed",
"generative"
],
"description": "GRIN verdict filter, derived from grinScores.extractive"
},
"editorialMode": {
"type": "string",
"enum": [
"grin",
"culture",
"research",
"ledger"
],
"description": "Filter to one editorial framework"
},
"plumbMin": {
"type": "number",
"description": "Minimum plumb (0-100)"
},
"plumbMax": {
"type": "number",
"description": "Maximum plumb (0-100)"
},
"since": {
"type": "string",
"description": "Only stories published at/after this ISO 8601 date"
},
"before": {
"type": "string",
"description": "Only stories published before this ISO 8601 date"
},
"cursor": {
"type": "string",
"description": "Opaque pagination cursor from a previous call"
},
"limit": {
"type": "number",
"description": "Max results (default 15, max 50)"
}
}
}🟢get_article(id, fields)
Fetch one BedrockNews story by id. Returns the transformed article plus its full structured analysis (GRIN scores, extraction graph, key stats, trajectory, what-to-watch). Pass fields:'grin' to get just the analysis payload without the heavy body text. Story ids contain slashes (e.g. 'ledger/2026/apr/06/iran-war-math-doesnt-work') — pass them verbatim.
入力スキーマ
{
"type": "object",
"properties": {
"id": {
"type": "string",
"description": "The story id, verbatim (ids contain slashes)"
},
"fields": {
"type": "string",
"description": "Optional projection: 'grin' for the analysis-only view"
}
},
"required": [
"id"
]
}🟢get_extraction_graph(id)
Return a story's extraction map as a real graph — nodes + edges data, not a picture. This is the load-bearing artifact for agents: re-render it in any style, or analyze who extracts value from whom. 'Not found' can mean the story has no graph (culture/research modes, or older transforms).
入力スキーマ
{
"type": "object",
"properties": {
"id": {
"type": "string",
"description": "The story id, verbatim (ids contain slashes)"
}
},
"required": [
"id"
]
}🟢get_related(id, limit)
Related BedrockNews stories for a given article — same section, ranked by editorial-mode match, plumb, source affinity, and recency.
入力スキーマ
{
"type": "object",
"properties": {
"id": {
"type": "string",
"description": "The source story id"
},
"limit": {
"type": "number",
"description": "Max related stories (default 4, max 12)"
}
},
"required": [
"id"
]
}🟢list_categories
The BedrockNews category taxonomy: ids (valid values for list_stories' category filter), display names, and which editorial framework each category runs (grin | culture | auto).
入力スキーマ
{
"type": "object",
"properties": {}
}🟢get_site_info
BedrockNews metadata: what the site is, how the GRIN/CLAIMS/NOVEL frameworks work, terms of use + attribution, and every machine-readable surface. Also returns the full capability map (every tool, prompt, and resource) — call this first.
入力スキーマ
{
"type": "object",
"properties": {}
}コミュニティ
エビデンス