codex-project-context

codex-project-context

Provides project-aware context for coding agents by selecting relevant AGENTS.md, docs, module paths, and decision records based on task, reducing redundant context loading via MCP stdio.

Category
Visit Server

README

Codex Project Context MCP

一个可迁移的全局 MCP Server + project-context 技能包。

它根据“项目路径 + 当前任务”选择需要读取的 AGENTS.md、docs、模块路径和决策记录,减少每次任务重复加载整套项目文档。

包含内容

server.py                         MCP stdio 入口
config/settings.json              规则与扫描配置
scanners/                         项目扫描、manifest、匹配、变更分析
context/                          Context Pack 聚合
skills/                           MCP 内置技能索引与推荐实现
skill/project-context/            Codex/兼容客户端的触发技能
install.ps1                       Windows 安装器
install.sh                        macOS/Linux 安装器
mcp-configs/                      Trae/Codex 配置示例
tests/                            无业务项目依赖的端到端测试
documentation/                    架构、变量、权限和测试说明

不包含:项目源码、项目索引、技能索引、虚拟环境、密钥、令牌和本机路径。

Windows 安装

git clone https://github.com/daodaobing/codex-project-context.git
cd codex-project-context
.\install.ps1

安装器会:

  1. 在 ~/.codex/context-server 安装 MCP 运行时;
  2. 创建独立 Python 虚拟环境并安装 requirements.txt;
  3. 将 project-context 安装到 ~/.codex/skills/project-context;
  4. 在 ~/.codex/config.toml 注册 context_server;
  5. 保留运行时生成的项目索引,不把索引上传到 Git。

需要指定 Python 时:

.\install.ps1 -Python "C:\Path\To\python.exe"

如果只想安装文件、不自动更新 Codex 配置:

.\install.ps1 -SkipConfig

安装后重启 Codex。验证:

codex mcp
& "$env:USERPROFILE\.codex\context-server\.venv\Scripts\python.exe" `
  "$env:USERPROFILE\.codex\context-server\tests\test_server.py"

macOS/Linux 安装

git clone https://github.com/daodaobing/codex-project-context.git
cd codex-project-context
./install.sh

也可以通过 CODEX_HOME 指定 Codex 配置根目录:

CODEX_HOME="$HOME/.codex" ./install.sh

Trae 使用

MCP Server 是跨客户端的稳定部分。把下面配置复制到 Trae 的 MCP 管理面板,并将 YOUR_USER 替换为当前用户名:

{
  "mcpServers": {
    "context_server": {
      "command": "C:\\Users\\YOUR_USER\\.codex\\context-server\\.venv\\Scripts\\python.exe",
      "args": [
        "C:\\Users\\YOUR_USER\\.codex\\context-server\\server.py"
      ]
    }
  }
}

完整示例见 mcp-configs/trae.windows.json。Trae 不同版本的配置入口和 SKILL.md 自动发现方式可能不同;如果版本支持项目级技能,将 skill/project-context 复制到项目的 .trae/skills/project-context,或通过 Trae 的技能/规则界面导入。若当前版本不识别 SKILL.md,将 trae/project-context.rules.md 复制到 Trae 的项目规则中即可。即使不安装技能,只配置 MCP 也可以直接使用所有 Context Tool。

使用方式

普通项目任务:

$project-context

项目路径:C:\Projects\MyProject

任务:修改登录页面并修复 token 过期问题。

技能会要求 Codex/兼容客户端先调用 get_project_context 或 get_context_pack,再使用 Serena 或客户端自身的代码导航能力定位实现。

新项目初始化:

$project-context

项目路径:C:\Projects\NewProject

请先预览初始化上下文文件,只读,不写入。

只有得到明确确认后,才调用 initialize_project_context(..., dry_run=false)。普通上下文加载不会创建或修改项目文件。

context-manifest.yaml

在项目根目录声明项目主动维护的上下文:

project:
  name: My Project
  type: code

modules:
  auth:
    paths:
      - src/core/auth
    documents:
      - docs/contracts/auth.md
    decisions:
      - docs/decisions.md
    keywords:
      - auth
      - login

knowledge_domains:
  billing_rules:
    documents:
      - docs/billing-prd.md
    keywords:
      - billing
      - 计费

modules 只放真实代码模块;没有代码路径的业务知识放 knowledge_domains。新增或修改 manifest 后,调用 scan_project(project_path, force=true) 刷新索引。

Tool

核心 Tool:

  • get_project_context
  • get_context_pack
  • scan_project
  • list_projects
  • get_project_index
  • initialize_project_context
  • suggest_projects
  • analyze_changes

技能推荐辅助 Tool:

  • refresh_skill_registry
  • list_skills
  • get_skill
  • recommend_skills

设计边界

  • 不替代 AGENTS.md、项目 docs 或 Serena;
  • 第一版规则匹配,不依赖向量数据库、Embedding 或 Agent 框架;
  • 索引只存路径、摘要、分类和模块关系,不存源码;
  • analyze_changes 只分析,不自动更新文档;
  • initialize_project_context 默认 dry-run,写入必须经过确认;
  • Server 通过 MCP stdio 运行,不需要公网服务或 API Key。

开发验证

& .\.venv\Scripts\python.exe tests\test_server.py

更详细的架构、变量、权限边界和测试证据见 documentation/。

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured