code-context-engine

概述

Code Context Engine 是一个基于 Python 的 MCP 服务器，它会创建代码库的可搜索索引，使 AI 编码助手能够检索相关代码片段，而不是读取整个文件。它在保持所有数据本地化的同时提供显著的代币节省（根据他们的基准为 94%）。该服务器提供 9 个 MCP 工具，用于上下文搜索、会话回溯和索引管理，可与多个 AI 编码环境一起使用，如 Claude Code、VS Code、Cursor 和 Gemini CLI。

试试问 AI

装完之后，这里有 5 个你可以让 AI 做的事：

• context_search

  Hybrid vector + BM25 search with graph expansion to find relevant code chunks

  试试:Find payment processing functions in the codebase试试:Show me how authentication is implemented试试:Search for functions that handle user registration
• expand_chunk

  Retrieve full source code for a compressed result

  试试:Show me the full implementation of that function试试:Expand the code chunk for payment processing试试:Get the complete source for the authentication module
• related_context

  Find code related via graph edges (calls, imports)

  试试:What files import the authentication module?试试:Show me functions called by payment processing试试:Find modules that use this database schema
• session_recall

  Recall decisions and context from past sessions

  试试:What decisions were made about authentication?试试:Show me previous work on payment processing试试:What were the reasons for choosing this architecture?
• record_decision

  Save a decision for future reference

  试试:Record decision: use JWT for authentication试试:Note that we're using Stripe for payment processing试试:Save the decision to use PostgreSQL as the database
• record_code_area

  Record which files were worked on for future reference

  试试:Record that we're working on the payment module试试:Note that we modified user authentication code试试:Remember that we're updating the database schema
• index_status

  Check the freshness and status of the code index

  试试:Is the code index up to date?试试:Show me the status of the index试试:Check if the index needs rebuilding
• reindex

  Re-index a specific file or the entire project

  试试:Re-index the payment processing module试试:Update the index for all modified files试试:Rebuild the entire project index
• set_output_compression

  Adjust response verbosity level

  试试:Set compression to standard试试:Use maximum compression for responses试试:Switch to lite mode for responses

可对比工具

安装

# 使用 uv 安装
uv tool install code-context-engine

# 或使用 pipx
pipx install code-context-engine

# 初始化你的项目
cd /path/to/your/project
cce init                              # 或者：cce init --agent all

{
  "mcpServers": {
    "code-context-engine": {
      "command": "uv",
      "args": ["run", "code-context-engine", "mcp"]
    }
  }
}

FAQ

CCE 支持哪些嵌入后端？

CCE 会自动检测最佳可用后端。当 Ollama 运行时，它使用 nomic-embed-text；或者当您安装 [local] 额外包时，包含 fastembed + ONNX Runtime 用于本地离线嵌入。

CCE 是否能同时与多个 AI 编码助手一起工作？

是的。CCE 可以对您的代码库进行一次索引，然后同时与 Claude Code、VS Code、Cursor、Gemini CLI、Codex CLI、OpenCode 和 Tabnine 一起工作，只需单个索引。

code-context-engine 对比