mcp-director
A management tool for Claude Code that optimizes performance by recommending and switching between task-specific MCP profiles. It enables users to dynamically update their configuration to load only the necessary MCP servers for their current session.
README
mcp-director
A lightweight MCP server that recommends and manages MCP profiles for Claude Code.
Problem
Claude Code loads all configured MCP servers at session start. With 20+ MCPs, startup is slow — but you only use 2-3 per session.
Solution
mcp-director stays as your only always-on MCP. Tell it what you want to do, and it recommends the right MCPs and switches your .mcp.json profile.
Install
Add mcp-director to your project's .mcp.json (create the file in your project root if it doesn't exist):
{
"mcpServers": {
"mcp-director": {
"command": "npx",
"args": ["-y", "mcp-director"]
}
}
}
Then start (or restart) your Claude Code session. mcp-director will be available as an MCP tool.
Note: No global install needed.
npxdownloads and runs it automatically.
Usage
Once Claude Code starts with mcp-director configured, just talk to Claude naturally:
1. Initialize — Import your existing MCPs
"Run the init tool"
This reads your current .mcp.json and imports all MCPs into .mcp-director/catalog.json so the director knows about them.
2. Get recommendations
"I want to create meeting notes"
Claude will call recommend and suggest relevant MCPs (e.g. whisper, notebooklm-mcp) along with matching profiles.
"I need to do some web research"
Use smart mode for LLM-assisted matching:
"Recommend MCPs for building a dashboard, use smart mode"
3. Switch profiles
"Apply the development profile"
This rewrites your .mcp.json to only include github, context7, and playwright. A backup is saved as .mcp.json.bak.
"Show me what the design profile would look like" (dry-run)
4. Restart session
Changes to .mcp.json take effect on the next Claude Code session start.
5. Create custom profiles
"Create a profile called 'data-work' with filesystem and sequential-thinking"
"Add exa to the research profile"
Tools
| Tool | Description |
|---|---|
recommend |
Suggest MCPs for a task (supports smart mode for LLM-assisted matching) |
apply_profile |
Switch .mcp.json to a profile (with backup and dry-run) |
list_profiles |
Show available profiles and current state |
create_profile |
Create a custom profile |
update_profile |
Add/remove MCPs from a profile |
delete_profile |
Remove a custom profile |
init |
Import existing .mcp.json into director catalog |
Profiles
Built-in profiles:
| Profile | MCPs |
|---|---|
| minimal | Director only |
| development | GitHub + Context7 + Playwright |
| research | Exa + Context7 + Sequential Thinking |
| meeting-notes | Whisper + NotebookLM + Filesystem |
| communication | Slack + Google Calendar + Linear |
| design | Figma + Playwright |
| automation | n8n + Filesystem |
| full | All MCPs enabled |
Create your own with create_profile.
Bundled Catalog
16 popular MCPs are included out of the box:
github playwright context7 memory filesystem sequential-thinking slack figma linear whisper notebooklm-mcp exa google-calendar screenpipe n8n-mcp
Add your own MCPs by running init or editing .mcp-director/catalog.json.
How It Works
- Bundled catalog of popular MCPs with keywords, categories, and capabilities
initimports your existing MCPs into a local.mcp-director/catalog.jsonrecommendscores MCPs against your task descriptionapply_profilerewrites.mcp.json(with.mcp.json.bakbackup)- Director always preserves itself in
.mcp.json
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。