Ai Context Inspector

io.github.cocaxcode由 GitHub 账号 cocaxcode 发布v0.4.9更新于 Oct 24, 2026

Scan, export, and import your AI ecosystem. Discover MCP servers, skills, and memories.

概览

AI 生成的概览

一个本地 MCP 服务器,扫描项目中的 AI 工具配置、MCP 服务器、技能、代理与记忆,并可在七种 AI 工具之间导出和导入。

可以做什么
扫描项目目录(可选包含用户级配置),发现 19 种工具的 AI 上下文文件、MCP 服务器配置、技能、代理和持久记忆,并可选实时连接 MCP 服务器列出其工具、资源和提示。根据扫描结果生成独立的 HTML 仪表板或原始 JSON 报告。将整个生态导出为可移植的打包文件,可按类别筛选,并选择脱敏、包含或按变量选择密钥。将打包文件导入目标工具(Claude、Cursor、Windsurf、Copilot、Gemini、Codex、OpenCode),自动适配路径和配置格式,默认以试运行方式执行。
适用场景
适合需要让助手盘点项目已有的 AI 工具与配置、比较不同工具的设置,或在 AI 工具之间迁移 MCP 服务器、规则、技能和代理的场景。
运行要求
通过 npx 以 Node.js 本地进程运行(stdio 传输,使用 --mcp 参数启动)。未声明需要账户、API 密钥或环境变量。仅在下载软件包以及启用实时探测时访问 HTTP/SSE MCP 服务器时需要网络。
安装前请注意
导入会向项目或用户目录写入配置文件,设置 force 时可能覆盖已有资源;默认是试运行,需明确确认后才会写入。导出可能包含密钥:secrets 选项可设为包含全部值或指定变量,分享打包文件前须谨慎选择脱敏方式。实时探测会启动配置中的 MCP 服务器命令并向其传递环境变量。扫描用户级配置会读取项目之外的文件。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Ai Context Inspector,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

@cocaxcode/ai-context-inspector

One command. Every AI config in your project. Instantly.
CLI + MCP server · 19 tools detected · Export/Import across 7 targets

[npm version] [npm downloads] [License] [Node] [19 AI tools]

Quick Start · MCP Server · What It Detects · Features · Tool Reference · Architecture


The most comprehensive AI ecosystem scanner available. AI Context Inspector discovers every AI configuration file, MCP server, custom skill, agent, and persistent memory in your project — across 19 tools — and gives you the full picture in seconds. Export your entire AI setup and import it into any of 7 supported tools (Claude, Cursor, Windsurf, Copilot, Gemini, Codex, OpenCode) with automatic secret redaction. Zero config. Zero runtime dependencies beyond MCP SDK and Zod. Everything runs locally — your configs and credentials never leave your machine.


Quick Start

bash
npx -y @cocaxcode/ai-context-inspector@latest

That's it. An HTML dashboard opens with everything your AI tools know about your project.

bash
# Scan a specific directorynpx -y @cocaxcode/ai-context-inspector@latest --dir /path/to/project
# Include user-level configs (~/.claude, ~/.gemini, etc.)npx -y @cocaxcode/ai-context-inspector@latest --user
# Live-connect to each MCP server and list its toolsnpx -y @cocaxcode/ai-context-inspector@latest --introspect
# Output raw JSON instead of HTMLnpx -y @cocaxcode/ai-context-inspector@latest --json
All CLI flags
FlagDescriptionDefault
--dir <path>Directory to scan. (current)
--userInclude user-level configs (~/)false
--introspectConnect to MCP servers and list toolsfalse
--timeout <ms>Introspection timeout per server10000
--jsonOutput JSON instead of HTMLfalse
--mcpRun as MCP server (stdio transport)false

MCP Server Setup

Run as an MCP server so your AI tools can inspect their own ecosystem.

Claude Code

bash
claude mcp add --scope user ai-context-inspector -- npx -y @cocaxcode/ai-context-inspector@latest --mcp

Claude Desktop

Add to claude_desktop_config.json:

json
{  "mcpServers": {    "ai-context-inspector": {      "command": "npx",      "args": ["-y", "@cocaxcode/ai-context-inspector@latest", "--mcp"]    }  }}
Cursor / Windsurf / VS Code / Codex CLI

Cursor / Windsurf — add to .cursor/mcp.json or equivalent:

json
{  "mcpServers": {    "ai-context-inspector": {      "command": "npx",      "args": ["-y", "@cocaxcode/ai-context-inspector@latest", "--mcp"]    }  }}

VS Code — add to .vscode/mcp.json:

json
{  "servers": {    "ai-context-inspector": {      "type": "stdio",      "command": "npx",      "args": ["-y", "@cocaxcode/ai-context-inspector@latest", "--mcp"]    }  }}

Codex CLI:

bash
codex mcp add ai-context-inspector -- npx -y @cocaxcode/ai-context-inspector@latest --mcp

Or add directly to any .mcp.json:

json
{  "mcpServers": {    "ai-context-inspector": {      "type": "stdio",      "command": "npx",      "args": ["-y", "@cocaxcode/ai-context-inspector@latest", "--mcp"]    }  }}

What It Detects

ToolContext FilesWhat's Found
Claude CodeCLAUDE.md, .claude/, .mcp.jsonInstructions, MCP config, skills, agents, memory
Cursor.cursorrules, .cursor/rules/, .cursorignoreRules, ignore patterns
Windsurf.windsurfrules, .windsurf/rules/Codeium/Windsurf rules
GitHub Copilot.github/copilot-instructions.md, .github/agents/Instructions, custom agents
Gemini CLIGEMINI.md, .gemini/, .geminiignoreConfig, rules
OpenAI CodexAGENTS.md, AGENT.md, .codex/Agent instructions
OpenCodeOPENCODE.md, .opencode/, opencode.jsonCLI config
Aider.aider.conf.yml, .aiderignoreConfig, model settings
Cline.clinerules, .clineignoreRules (file or directory)
Roo Code.roo/rules/, .roorules, .rooignoreRules
Continue.dev.continuerules, .continue/config.yamlRules, config
Amazon Q.amazonq/rules/Developer rules
Augment.augment/rules/, .augment-guidelinesCode rules
Replit.replit.mdAgent instructions
Firebase Studio.idx/airules.mdAI rules
VS Code.vscode/mcp.jsonMCP configuration
Tabnine.tabnine.yamlAI config
Sourcegraph.sourcegraph/Cody config
UniversalCONVENTIONS.mdMulti-tool conventions

Tip: Pass --user to also scan user-level configs: ~/.claude/, ~/.gemini/, ~/.codex/, ~/.continue/, ~/.aider.conf.yml, ~/.augment/, ~/.github/agents/, ~/.codeium/, ~/.opencode/, ~/.tabnine/.

Beyond context files, the scanner also detects:

  • MCP servers from .mcp.json, ~/.claude.json, .vscode/mcp.json, Claude Desktop config — with optional live introspection
  • Custom skills in .claude/skills/ (including symlinks from npx skills add)
  • Custom agents in .claude/agents/ with memory detection
  • Persistent memories — Claude Memory, agent memory, engram, OpenSpec, ATL

Features

Scan

bash
npx -y @cocaxcode/ai-context-inspector@latestnpx -y @cocaxcode/ai-context-inspector@latest --user --introspect

Generates a self-contained HTML dashboard with dark/light mode, search, collapsible sections, file previews, and color-coded tool badges. Pass --json for raw output.

Export

bash
# Export your AI ecosystem to .aci/bundle.jsonnpx -y @cocaxcode/ai-context-inspector@latest export
# Export only MCP and context configs, redact secretsnpx -y @cocaxcode/ai-context-inspector@latest export --only mcp,context --secrets none
Export flags
FlagDescriptionDefault
--dir <path>Directory to export.
--output <path>Output directory.aci/
--include-userInclude user-level configsfalse
--only <categories>Filter: mcp,skills,agents,memories,contextall
--secrets <mode>none (redact) or all (include)interactive

Import

bash
# Import into a Cursor projectnpx -y @cocaxcode/ai-context-inspector@latest import --target cursor
# Auto-detect tool, skip confirmation, overwrite existingnpx -y @cocaxcode/ai-context-inspector@latest import --yes --force
Import flags
FlagDescriptionDefault
[file]Path to bundle JSONauto-detect .aci/bundle.json
--dir <path>Target directory.
--target <tool>Target tool (see table)auto-detect
--scope <scope>project or userper-resource
--forceOverwrite existing resourcesfalse
--yesSkip confirmation promptfalse
--only <categories>Filter: mcp,skills,agents,memories,contextall
--secrets <mode>none or allinteractive

Supported Import Targets

TargetMCP ConfigContext FileRules DirSkillsAgents
claude.mcp.jsonCLAUDE.md--.claude/skills/.claude/agents/
cursor.cursor/mcp.json.cursorrules.cursor/rules/----
windsurf.mcp.json.windsurfrules.windsurf/rules/----
copilot.vscode/mcp.json.github/copilot-instructions.md.github/instructions/--.github/agents/
gemini.gemini/settings.jsonGEMINI.md.gemini/rules/----
codex.mcp.jsonAGENTS.md.codex/rules/----
opencodeopencode.jsonOPENCODE.md.opencode/rules/----

Note: Secrets handling has three modes: none (redact all — safe for sharing), all (include as-is), or interactive (prompt per variable, CLI default). Environment variables in MCP configs are detected automatically.


MCP Tool Reference

When running as an MCP server (--mcp), five tools are exposed:

scan

Discover the complete AI ecosystem in a project directory.

ParameterTypeRequiredDescription
directorystringNoPath to scan (default: cwd)
include_userbooleanNoInclude user-level configs
introspectbooleanNoLive-connect to MCP servers
timeoutnumberNoIntrospection timeout (ms)

introspect_mcp

Connect to a specific MCP server and list its tools, resources, and prompts.

ParameterTypeRequiredDescription
commandstringYes*Command to launch the server
argsstring[]NoArguments for the command
envobjectNoEnvironment variables
urlstringYes*URL for HTTP/SSE servers
transportstringNostdio (default) or http
timeoutnumberNoConnection timeout (ms)

*Either command (stdio) or url (http) is required.

generate_report

Generate a standalone HTML dashboard from scan results.

ParameterTypeRequiredDescription
directorystringNoPath to scan
include_userbooleanNoInclude user-level configs
introspectbooleanNoIntrospect MCP servers
timeoutnumberNoIntrospection timeout (ms)

export_ecosystem

Export the complete AI ecosystem to a portable .aci/bundle.json.

ParameterTypeRequiredDescription
dirstringNoDirectory to scan (default: cwd)
include_userbooleanNoInclude user-level configs
onlystring[]NoCategories: mcp, skills, agents, memories, context
secretsstringNo"none" (default), "all", or ["VAR1", "VAR2"]

import_ecosystem

Import a bundle into a project, adapting configuration to the target tool.

ParameterTypeRequiredDescription
filestringNoPath to bundle (auto-detects .aci/bundle.json)
dirstringNoTarget directory (default: cwd)
targetstringNoTarget tool (auto-detects): claude, cursor, windsurf, copilot, gemini, codex, opencode
scopestringNoproject or user
forcebooleanNoOverwrite existing resources
confirmbooleanNoExecute import (default: false = dry-run)
onlystring[]NoCategories to import
secretsstringNo"none", "all", ["VAR1"], or {"VAR1": "value"}

Warning: import_ecosystem defaults to dry-run mode (confirm: false). Set confirm: true to actually write files.


Architecture

src/├── index.ts              # Entry: CLI vs MCP mode routing├── cli.ts                # CLI arg parsing + orchestration├── server.ts             # createServer() MCP factory├── scanner/              # 6 parallel scanners (Promise.all)│   ├── catalog.ts        # AI_FILE_CATALOG — 50+ entries, 19 tools│   ├── context-files.ts  # Scan .md, rules, configs│   ├── mcp-configs.ts    # Parse .mcp.json, ~/.claude.json, etc.│   ├── mcp-introspect.ts # Connect to servers, list tools/resources│   ├── skills.ts         # Skills + symlinks + frontmatter│   ├── agents.ts         # Agents + memory detection│   └── memories.ts       # engram, openspec, .atl, claude memory├── ecosystem/            # Export/import engine│   ├── export.ts         # Scan → bundle → .aci/bundle.json│   ├── import.ts         # Load → plan → execute│   ├── target-map.ts     # Path configs for 7 AI tools│   ├── detect-target.ts  # Auto-detect tool in project│   └── secrets.ts        # Env var detection + redaction├── report/               # HTML dashboard generator│   ├── generator.ts      # generateHtml(ScanResult) → string│   ├── sections.ts       # Section renderers│   ├── styles.ts         # CSS (dark/light)│   └── scripts.ts        # JS (collapse, search, preview)└── tools/                # 5 MCP tool handlers    ├── scan.ts    ├── introspect.ts    ├── report.ts    ├── export.ts    └── import.ts

Stack: TypeScript 5 (strict, ESM) / MCP SDK / Zod / tsup / Vitest


MIT · Built by cocaxcode

来源:README.md,提交 2e12a55

工具

0
SourceWeft 尚未在沙箱中启动过该软件包,因此其工具未知。

版本历史

1
  1. v0.4.9最新Sep 16, 2026

评分

已满足适用检查项的 50%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通25 分

    请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。

  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分