kube-mcp-operator
Exposes Kubernetes microservice APIs as MCP services by injecting a sidecar into annotated deployments.
README
Kube MCP Operator
This repository provides a minimal implementation of a Kubernetes operator and sidecar that expose microservice APIs as MCP services.
Components
- Sidecar - Lightweight FastAPI application that proxies requests to the main container and exposes its OpenAPI specification from a configurable path.
- Operator - A Kopf based operator watching for deployments annotated with
mcp-server: "true", injecting the sidecar and creating a service for it. - CRD -
MCPConfigallows selecting deployments by label and exposing only those annotated. - Helm chart - Installs the operator and CRD.
Architecture
graph TD
subgraph Pod
App[Microservice]
Sidecar[MCP Sidecar]
App <--> Sidecar
end
Client --> Sidecar
Operator[MCP Operator] -- watches --> Deployment["Deployment with mcp-server=true"]
Operator -- injects sidecar --> Deployment
Operator -- creates --> Service
Service --> Sidecar
Building
Sidecar image:
docker build -t mcp-sidecar ./sidecar
Operator image:
docker build -t mcp-operator ./mcp_operator
Installing with Helm
helm install mcp-operator charts/mcp-operator
Usage
Annotate your deployment to enable MCP integration:
metadata:
annotations:
mcp-server: "true"
When the operator sees this annotation it injects the MCP sidecar and creates a service <deployment>-mcp exposing port 8000.
Existing annotated deployments are processed on startup so sidecars are injected even if the operator is installed later.
Configuration
The sidecar locates the upstream service using environment variables:
SERVICE_HOST– hostname of the microservice (defaultlocalhost)SERVICE_PORT– port of the microservice (default80)OPENAPI_PATH– path to the OpenAPI or Swagger file including the file name (defaultopenapi.json)
Requests to /openapi.json on the sidecar return the file from
http://$SERVICE_HOST:$SERVICE_PORT/$OPENAPI_PATH.
Testing
Run the unit tests with coverage:
pip install -e .[dev]
pytest --cov=sidecar --cov=mcp_operator --cov-report=term --cov-fail-under=80
CI Artifacts
The GitHub Actions workflow builds the sidecar Docker image and packages the
Helm chart on each push. When changes land on main, a release is created using
python-semantic-release (GitHub Action v9) which automatically bumps the version, updates the
changelog and attaches:
mcp-sidecar.tar– the Docker image saved as a tarballmcp-operator-<version>.tgz– the packaged Helm chart
Version history lives in CHANGELOG.md and is maintained by
python-semantic-release via the latest GitHub Action.
Security
Each build scans the sidecar image using Trivy. The generated report is available in SECURITY.md and is committed back to the repository whenever a new release is published.
Contributing
Please see CONTRIBUTING.md for guidelines on developing and submitting changes.
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。