Context Pill

Daily feed of AI releases, tools and techniques, each labeled with how it was checked.

我该使用它吗

质量与安全性

A
描述质量
100%
模式完整度
93%
命名质量
100%
投毒风险
100%
权限匹配度
100%
协议合规性
100%

基于对工具定义和协议合规性的自动分析。

上下文开销

~470token 数(工具定义)
~656 B典型响应大小
对注意力的影响极小(占 128k 上下文窗口的 0.37%)

这是每次将服务器的工具加载到模型上下文窗口时所消耗的大致 token 数。数值越高,可用于其他任务的注意力就越少。

安装

一键安装

将以下内容添加到你的 `claude_desktop_config.json` 文件中:

{
  "mcpServers": {
    "context-pill": {
      "url": "https://contextpill.com/mcp"
    }
  }
}

远程端点

https://contextpill.com/mcpstreamable-http

它能做什么

工具清单

工具(3)

🟢 只读🟡 写入🔴 删除⚪ 未知
🟢get_latest(topic, limit)

Return today's checked AI claims (releases, tools, techniques), each with its verification status, source trust score and evidence links. Optionally filter by a topic word such as 'claude', 'codex' or 'open weights'. Call this first when the user asks what's new in AI.

输入模式

{
  "type": "object",
  "properties": {
    "topic": {
      "type": "string",
      "description": "Optional word or phrase to filter items by"
    },
    "limit": {
      "type": "integer",
      "minimum": 1,
      "maximum": 30,
      "description": "Maximum items to return (default 15)"
    }
  },
  "additionalProperties": false
}
🟢search_claims(query, limit)

Search the claims Context Pill has published (the ledger), with their status at publication, how each was checked, current state (pending, resolved, corrected or closed-inconclusive), the re-check outcome and any correction. Removed items keep only an id, date and reason, so they don't show up in a word search. Use it to see whether a specific tool, model or feature claim has been checked. Call this first when the user asks whether Context Pill has checked a specific tool, model or claim.

输入模式

{
  "type": "object",
  "properties": {
    "query": {
      "type": "string",
      "description": "Words to look for, e.g. a tool or model name"
    },
    "limit": {
      "type": "integer",
      "minimum": 1,
      "maximum": 30,
      "description": "Maximum claims to return (default 10)"
    }
  },
  "required": [
    "query"
  ],
  "additionalProperties": false
}
🟢list_sources(name)

List the sources Context Pill reads, with tier, trust score (posterior mean), its conservative lower bound (LB10) and the number of resolved tips. Optionally filter by name. Use it when the user asks how far to trust a source.

输入模式

{
  "type": "object",
  "properties": {
    "name": {
      "type": "string",
      "description": "Optional source name filter"
    }
  },
  "additionalProperties": false
}

社区

评价此服务器

证据

最近观测

已验证未记录版本3 个工具