
CodeScope MCP 为大型语言模型和人工智能助手提供可控、智能且上下文高效的本地代码库访问。
概览
CodeScope MCP 为大型语言模型和人工智能助手提供可控、智能且上下文高效的本地代码库访问。 CodeScope 不再天真地倾倒整个文件夹或用垃圾文件(node_modules、.venv、二进制文件、锁文件)淹没上下文窗口,而是作为安全和上下文防火墙,提供分层工具,支持代码库探索、代码搜索和文件检索。 在使用基于MCP的AI编码助手或LLM聊天代理时,授予原始文件系统访问权限通常会导致:1. : 巨大的文件夹如node_modules/、.git/、dist/或.venv/泛滥提示令牌。2. : 读取多兆字节文件或大型锁文件(package-lock.json、poetry.lock)。3. : 非意图将二进制或编译文件(.pyc、图片、.exe、.so)转储到文本提示中。4. : 路径穿越攻击(例如../../../../etc/passwd 或访问项目根以外的敏感文件)。 弥合了这一鸿沟。
概述(AI 翻译)
智能且代币优化的代码库上下文MCP服务器,专为云聊天大型语言模型(Claude、ChatGPT、Gemini)构建。
CodeScope MCP 通过 Model Context 协议(MCP) 通过 可流式 HTTP 传输,为基于云的聊天大型语言模型应用提供受控、智能且上下文高效的本地代码库访问。
CodeScope 不再天真地倾倒整个文件夹或用垃圾文件(node_modules、.venv、二进制文件、锁文件)淹没上下文窗口,而是作为安全和上下文防火墙,提供分层工具包,支持代码库探索、代码搜索和文件检索。
📑 目录
- 这是关于什么的
- 目标受众:为云聊天大型语言模型(非 AI IDE)设计
- 为什么你应该使用它
- 它的功能(核心工具)
- 架构概述
- 系统流程与分层发现
- 项目结构
- 入门指南
- 先决条件
- 安装与设置
- 环境配置
- 配置与尺寸指南
- 运行服务器
- 使用写作工具与人工智能提示
- 用ngrok曝光(远程/云设置)
- 连接MCP客户端
- 远程连接器指南(Claude、ChatGPT、Gemini)
本页由 AI 翻译,技术细节请以原文为准。
README
以下为原文 README 节选,完整内容请查看原文。
🔭 CodeScope MCP
Intelligent, Token-Optimized Codebase Context MCP Server Built for Cloud Chat LLMs (Claude, ChatGPT, Gemini).
CodeScope MCP provides cloud-based Chat LLM applications with controlled, intelligent, and context-efficient access to local codebases over the Model Context Protocol (MCP) using Streamable HTTP transport.
Instead of naively dumping entire folders or overwhelming context windows with garbage files (node_modules, .venv, binaries, lock files), CodeScope acts as a security and context firewall, offering a layered toolset for codebase exploration, code search, and file retrieval.
📑 Table of Contents
- What This Is About
- Target Audience: Built for Cloud Chat LLMs (Not for AI IDEs)
- Why You Should Use It
- What It Does (Core Tools)
- Architecture Overview
- System Flow & Layered Discovery
- Project Structure
- Getting Started
- Working with Write Tools & AI Prompting
- Automated Verification & .codescope.toml Guide
- Exposing with ngrok (Remote / Cloud Setup)
- Connecting to MCP Clients
- Cross-Platform Support (macOS, Linux & Windows)
- Security & Constraints
- Running Tests
- License
💡 What This Is About
When using AI coding assistants or LLM chat agents over MCP, granting raw filesystem access often leads to:
- Context Window Blowup: Giant folders like
node_modules/,.git/,dist/, or.venv/flooding prompt tokens. - High Token Costs & Slow Responses: Reading multi-megabyte files or large lock files (
package-lock.json,poetry.lock). - Binary Corruption: Unintended dumping of binary or compiled files (
.pyc, images,.exe,.so) into text prompts. - Security Vulnerabilities: Path traversal attacks (e.g.
../../../../etc/passwdor accessing sensitive files outside project roots).
CodeScope MCP bridges this gap. It provides a standardized MCP server built on FastMCP with Streamable HTTP that delivers smart, token-efficient, .gitignore-aware file reading, fast code/regex search, and metadata analysis.
🎯 Target Audience: Built for Cloud Chat LLMs (Not for AI IDEs)
安装
This server does not publish a one-line install command.
Open the repository installation guide 新手?先看 mcpskill.org 教程 →配置
{
"mcpServers": {
"codescope": {
"url": "http://127.0.0.1:8000/mcp",
"transport": "streamable-http"
}
}
}