n8n Local Control
MCP server that turns natural-language requests into configured n8n workflows and safely operates a local n8n Docker stack, including workflow management, API access, and Docker lifecycle operations with confirmation-based safety.
README
n8n Local Control
Turn natural-language requests into configured n8n workflows and safely operate a local n8n Docker stack from compatible AI agents.
This repository is dual packaged:
plugin.jsonandmcp.jsonimplement the portable Agent Plugins 1.0 format..codex-plugin/plugin.jsonand.mcp.jsonretain native Codex packaging.skills/contains portable Agent Skills.
Capabilities
- Search n8n nodes and templates, construct workflows, validate nodes and complete workflows, inspect executions, and manage credentials through pinned
n8n-mcp@2.68.0. - Discover and call the live n8n Public API with exact OpenAPI allowlisting.
- Use expiring two-phase confirmation tokens for API writes and stack mutations.
- Inspect health and logs; start, stop, restart, back up, restore, and upgrade the exact Docker Compose deployment discovered from container labels.
- Create consistent stopped-volume backups with SHA-256 metadata and restore/upgrade rollback.
Requirements
- Node.js 18 or newer.
- n8n available at
http://localhost:5678by default. - For full Windows stack control: PowerShell and Docker Desktop with an n8n Compose deployment.
- An n8n API key is optional for documentation and static validation, but required for workflow/API management.
Agent Plugins installation
Clone the repository and load its directory with an Agent Plugins 1.0-compatible client. The portable MCP entries use ${PLUGIN_ROOT} and ${PLUGIN_DATA} exactly as defined by the specification.
git clone https://github.com/RobertIonutF/n8n-local-control.git
The package is Windows-first for DPAPI and Docker administration. On other platforms the portable skills and MCP workflow engine can still load, but API-key storage must be supplied by the client or process environment and Windows stack tools are unavailable.
One-time Windows setup
-
In n8n, open Settings -> n8n API and create an API key with only the scopes you want the agent to use.
-
Run:
& '.\scripts\setup.ps1' -
Paste the key into the hidden terminal prompt. The script validates it and stores only a Windows DPAPI-encrypted value.
The setup script discovers the compose project, service, compose file, and n8n data volume from Docker labels. Override discovery only when needed:
& '.\scripts\setup.ps1' -Container n8n -BaseUrl 'http://localhost:5678' -ComposeFile 'D:\path\to\docker-compose.yml'
When an Agent Plugins client supplies PLUGIN_DATA, run setup from a terminal with that environment value or pass -StateDirectory explicitly so the launcher and setup helper share the same state directory. Codex falls back to %LOCALAPPDATA%\Codex\n8n-local-control.
Never paste the n8n API key or third-party credential secrets into chat.
Safety model
Read-only inspection, workflow design, inactive creation, and static validation can run autonomously. Activation, publication, side-effectful execution, deletion, credential changes, users/security, community packages, service interruption, restore, and upgrade require explicit confirmation. The local MCP server also requires an expiring prepared-operation token before mutations.
Validation
python .\scripts\validate_agent_plugin.py
node .\tests\server.test.mjs
node .\tests\mcp-handshake.test.mjs
powershell -NoProfile -ExecutionPolicy Bypass -File .\tests\dpapi.test.ps1
Codex plugin authors can additionally run the plugin-creator validator from their local Codex installation.
Live Docker tests are intentionally separate because they briefly stop n8n or create disposable Docker resources.
Privacy
The plugin does not embed credentials. The upstream n8n-MCP telemetry is disabled by the launcher. Sensitive API response fields are redacted, and Docker/API targets are restricted before execution.
License
MIT. See LICENSE.
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。