context-engineering

概述

这是一个全面的 MCP 服务器，实现了 WARNERCO Schematica 应用程序，展示了生产级别的语义记忆系统。它利用 FastMCP、FastAPI 和 LangGraph 实现了所有四个 CoALA 内存层级（工作记忆、情节记忆、语义记忆、程序记忆），并通过 9 节点管道进行处理。该服务器提供 28 个 MCP 工具、11 个资源和 5 个提示，用于构建具有持久记忆的 AI 系统。它具备渐进式工具加载功能，并使用 Park 等人的评分公式进行情节记忆检索。

试试问 AI

装完之后，这里有 5 个你可以让 AI 做的事：

• warn_search_tools

  Keyword discovery of MCP tools with detail levels

  试试:Search for tools related to memory management试试:Find tools with 'episodic' in the name
• warn_describe_tool

  Get full schema for a specific MCP tool by name

  试试:Describe the retrieve tool试试:Show me the details of log_episode tool
• parse_intent

  Parse user intent for the LangGraph pipeline

  试试:I need to find information about vector stores试试:Help me understand CoALA memory tiers
• query_graph

  Query the knowledge graph for relationships and entities

  试试:What entities are related to vector stores?试试:Show relationships between memory models
• inject_scratchpad

  Inject working memory into the current session

  试试:Add this observation to my working memory试试:Record that I'm focusing on memory consolidation
• recall_episodes

  Recall episodic memories based on relevance and importance

  试试:What did we discuss about semantic memory last week?试试:Find important episodes about RAG systems
• retrieve

  Retrieve information from semantic memory vector store

  试试:Retrieve facts about long-term memory in AI试试:Find information about context engineering
• compress_context

  Compress retrieved context for efficient processing

  试试:Compress this information about vector databases试试:Summarize the key points about memory consolidation
• reason

  Reason over compressed context to generate insights

  试试:Reason about the trade-offs between different memory systems试试:Analyze the impact of context length on AI performance
• respond

  Generate a response based on the processed context

  试试:Explain how CoALA memory tiers work试试:Summarize what we've learned about MCP tools
• log_episode

  Log the current interaction as an episodic memory

  试试:Log this conversation as an important episode试试:Record this interaction with high importance
• warn_episodic_recall

  Recall episodic memories with detailed scoring breakdown

  试试:Show me the scoring breakdown for memory recall试试:Why did certain memories surface higher than others?

说明：Tool names inferred from the LangGraph pipeline description and CoALA memory tiers documentation. Descriptions are based on their roles in the pipeline and memory system. Signatures are not explicitly provided in the README.

可对比工具

安装

git clone https://github.com/timothywarner-org/context-engineering.git
cd context-engineering/labs/lab-01-hello-mcp/starter
npm install && npm start

cd src/warnerco/backend
uv sync
uv run uvicorn app.main:app --reload    # HTTP 服务器
uv run warnerco-mcp                      # MCP stdio 服务器

{
  "mcpServers": {
    "warnerco": {
      "command": "uv",
      "args": ["run", "warnerco-mcp"],
      "cwd": "C:/github/context-engineering/src/warnerco/backend"
    }
  }
}

FAQ

CoALA 内存系统是什么？

CoALA（上下文感知长期架构）是一个四层记忆系统，包含工作记忆、情节记忆、语义记忆和程序记忆，在此 MCP 服务器中实现。

如何测试此 MCP 服务器？

您可以使用 MCP Inspector 测试服务器，命令为：`npx @modelcontextprotocol/inspector uv run warnerco-mcp`，它将在 http://localhost:5173 打开。

context-engineering 对比