memtrace-public

概述

Memtrace 将代码库转化为实时知识图谱，使 AI 编码代理能够在毫秒级时间内进行查询。它维护每个会话中的函数、类、调用边和版本的结构化记忆，使代理能够导航复杂的代码库，而无需重新读取文件或破坏不可见的依赖。该系统使用 Tree-sitter AST 解析在本地构建确定性符号关系，避免昂贵的 LLM 推理，同时保持代码分析的高准确性。

试试问 AI

装完之后，这里有 5 个你可以让 AI 做的事：

• find_code

  Hybrid search using BM25 + semantic + RRF to find code in the repository

  试试:Find functions that handle user authentication试试:Search for all references to payment processing
• find_symbol

  Exact or fuzzy search for symbols (functions, classes, etc.) with Levenshtein support

  试试:Find the User class definition试试:Locate the process_data function
• analyze_relationships

  Analyze relationships between symbols including callers, callees, hierarchy, and imports

  试试:Show what calls the calculate_totals function试试:Find all imports in the auth module
• get_symbol_context

  Get a 360-degree view of a symbol with all its relationships and dependencies

  试试:Show me everything about the PaymentProcessor class试试:Get full context for the handle_error function
• get_impact

  Analyze the impact of changing a symbol with risk rating

  试试:What will break if I change the database connection试试:Show the blast radius of modifying user authentication
• detect_changes

  Map code changes to affected symbols using diff-based analysis

  试试:What symbols are affected by recent changes试试:Show which files need updating after the API refactor
• find_dead_code

  Detect code that has no callers and is potentially unused

  试试:Find functions that are never called试试:Identify unused classes in the codebase
• find_most_complex_functions

  Identify functions with the highest cyclomatic complexity

  试试:Find the most complex functions in the project试试:Show complexity hotspots in the payment module
• get_evolution

  Analyze how symbols have evolved over time using 6 different scoring modes

  试试:How has the user model evolved试试:Show the most impactful changes this week
• get_timeline

  Get the full version history of a symbol

  试试:Show the timeline of the authentication system试试:When was the payment processing last refactored
• find_bridge_symbols

  Find symbols with high betweenness centrality that connect different parts of the graph

  试试:Find bridge symbols in the codebase试试:Identify critical connectors between modules
• list_communities

  Identify architectural modules using the Louvain community detection algorithm

  试试:Show the architectural modules in the codebase试试:Find communities related to user management

可对比工具

安装

{
  "mcpServers": {
    "memtrace": {
      "command": "memtrace",
      "args": ["serve"]
    }
  }
}

FAQ

Memtrace 与传统代码搜索工具有何不同？

Memtrace 创建双时态知识图谱，维护代码元素随时间推移的历史上下文和关系，而传统搜索工具仅提供当前状态。

Memtrace 支持哪些编程语言？

Memtrace 通过 Tree-sitter 解析器支持 20+ 种编程语言，包括主流语言如 Python、JavaScript、Java、C++，以及具有专用扫描器的特定语言。

memtrace-public 对比