ContextOS

概述

ContextOS 是一个 MCP 服务器，旨在成为 AI 上下文的操作系统层，通过整合、扩展和超越七个领先开源仓库的能力而构建。它提供包含六个认知原语的推理层，根据数据变化率处理检索的路由器，以及自愈索引的生命周期管理器。平台包含语义意图路由、请求追踪、模式注册表和多工作区认证等编排功能。

试试问 AI

装完之后，这里有 5 个你可以让 AI 做的事：

• register_source

  Register a data source with specific churn profile and indexing strategy

  试试:Register a new data source for search queries with live churn class试试:Add a cold data source with vector indexing strategy for policy documents
• router

  Access the data source router for managing and retrieving information

  试试:Show me the registered data sources and their churn profiles试试:Retrieve information using the churn-aware routing system
• semantic_intent_router

  Classify incoming requests and dispatch to correct processing layer

  试试:Classify this request and route it appropriately试试:Determine if this task requires retrieval or synthesis
• request_tracing

  Track lineage and observability data for tool calls

  试试:Show me the tracing information for the last tool call试试:Identify which layers fired during processing
• active_forgetting

  Filter out retrieved context that degrades output quality

  试试:Apply active forgetting to this retrieved information试试:Remove noise from the context before processing
• reasoning_depth_calibration

  Determine appropriate depth of thinking for a given problem

  试试:Calibrate reasoning depth for this complex analysis试试:Estimate how much thinking this problem deserves
• synthesis_detection

  Identify whether the task requires retrieval or reasoning

  试试:Detect if this is a synthesis or retrieval task试试:Determine if more data would help or hurt this analysis
• unknown_unknown_sensing

  Detect when missing entire categories of information

  试试:Check for unknown unknowns in this analysis试试:Flag potential missing information categories
• productive_contradiction

  Handle conflicting data as useful signal rather than noise

  试试:Resolve contradiction between these data points试试:Analyze conflicting signals to find insights
• context_dependent_gravity

  Re-weight memory importance based on current question

  试试:Apply context-dependent gravity to this memory试试:Re-weight these memories based on the current query
• rebuild_index

  Trigger rebuild of data indexes manually or automatically

  试试:Rebuild the index for the keyword data source试试:Check if indexes need rebuilding and update them
• embedding_drift_detection

  Detect and handle embedding model version mismatches

  试试:Check for embedding model drift in all indexes试试:Update indexes after embedding model change

说明：Tool names were inferred from feature descriptions and code examples in the README. While not explicitly listed as a tool section, the functionality descriptions and code examples clearly indicate these MCP tools.

可对比工具

安装

pip install contextos

{
  "mcpServers": {
    "contextos": {
      "command": "python",
      "args": ["-m", "contextos"]
    }
  }
}

FAQ

ContextOS 与其他 MCP 服务器的包装器有何不同？

ContextOS 不是包装器，而是一个平台，将其他工具转换为在其上运行的模块，提供额外的编排和认知能力。

六个认知原语是什么？

主动遗忘、推理深度校准、综合检测、未知未知感知、有效矛盾和上下文依赖的权重调整。

ContextOS 对比