MCP Control Plane
Open-source, Windows-first control plane for operating local and homelab services through typed MCP servers, featuring 13 servers and 290 tools with infrastructure adapters for Windows, Ubuntu, Docker, and more.
README
MCP Control Plane
Open-source, Windows-first control plane for operating local and homelab services through typed Model Context Protocol servers.
The repository combines MCP 2.0 stateless servers, durable worker orchestration, Context Store memory, security-aware infrastructure adapters, Agent Skills, and optional OpenTelemetry export.
Highlights
- 13 MCP servers and 290 narrowly scoped tools.
- JSON Schema 2020-12 output contracts and structured tool results.
- MCP Apps support with a text fallback for hosts without embedded UI.
- Durable internal Tasks pilot with explicit wire-capability detection.
- Windows, Ubuntu, Docker, MikroTik, Home Assistant, observability, mail, Steam Deck, and ComfyUI integrations.
- Stateless MCP conformance suite and offline Windows CI.
- Project-scoped MCP security and control-plane audit skills.
Architecture
flowchart LR
C["Codex / MCP client"] --> R["MCP 2.0 server entrypoints"]
R --> S["Shared runtime and typed contracts"]
S --> W["Durable workers"]
S --> M["Context Store"]
R --> I["Infrastructure adapters"]
S --> O["Optional OpenTelemetry"]
I --> H["Windows / Ubuntu / Docker / Router / Home Assistant"]
Entrypoints remain at the repository root because Codex configuration and local automation reference those paths directly. Shared behavior lives in:
mcp_runtime.py— server factory and middleware wiring.mcp_contracts.py— closed typed output models.mcp_extensions.py— MCP Apps and durable Tasks pilot.mcp_observability.py— opt-in telemetry configuration.codex_workers.pyandworkers_mcp.py— bounded parallel execution.codex_memory.pyandmemory_mcp.py— durable Context Store.
See Architecture for the subsystem map.
Quick start
Requirements:
- Windows 11 or Windows Server
- Python 3.12
- PowerShell 7 recommended
- GitHub CLI for Agent Skill management
python -m venv .venv
.\.venv\Scripts\python.exe -m pip install -r requirements.txt
Copy-Item .env.example .env
Edit .env locally. Never commit it. Start an individual server with:
.\.venv\Scripts\python.exe workers_mcp.py
Use config.example.toml as a registration template and
copy only the servers you need into the Codex MCP configuration. Long-lived MCP
processes must be restarted after code or dependency changes.
Validation
.\.venv\Scripts\python.exe -m pip install -r requirements-dev.txt
.\.venv\Scripts\python.exe -m pytest
.\.venv\Scripts\python.exe skills\mcp-control-plane-audit\scripts\audit_control_plane.py --root . --fail-on medium
powershell -NoProfile -ExecutionPolicy Bypass -File .\scripts\run_mcp_conformance.ps1
Live infrastructure smoke tests are opt-in and are skipped in normal CI.
Agent Skills
Install the repository-owned audit skill project-scoped:
gh skill install . mcp-control-plane-audit --from-local --agent codex --scope project
Remote skills should be reviewed and installed at an immutable commit:
gh skill preview OWNER/REPO PATH@COMMIT
gh skill install OWNER/REPO PATH --agent codex --scope project --pin COMMIT
Security
This repository contains control-plane code, not credentials. Runtime secrets
belong in .env, the OS credential store, or a dedicated secret manager.
Mutation tools must enforce authorization inside the MCP/API implementation;
agent prompts and UI confirmations are not security boundaries.
Read SECURITY.md before connecting production infrastructure.
Repository scope
Large captures, firmware, generated artifacts, operational snapshots, vendored tools, local databases, private keys, and runtime state are intentionally excluded. Security Session Gate, Shadow RDP Console, branding assets, and firmware research are maintained as separate projects.
License
Released under the permissive MIT 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。