Kubernetes MCP Server – 5G Core Edition
Exposes Kubernetes cluster state with specialized telecom awareness of 5G Core network functions and topologies to MCP-compatible LLMs. It enables natural language analysis of 5G workloads, network slices, UPF data planes, and cluster health.
README
Kubernetes MCP Server – 5G Core Edition
An MCP (Model Context Protocol) server that exposes your Kubernetes cluster state with full 5G Core / telecom awareness to any MCP-compatible LLM (Claude Desktop, etc.).
Features
Generic Kubernetes tools
| Tool | Description |
|---|---|
k8s_get_pods |
List pods with optional NF-type filter |
k8s_get_deployments |
Deployment replicas and health |
k8s_get_services |
Services + SBI endpoint detection |
k8s_get_pod_logs |
Logs with 5G error annotations |
k8s_describe_pod |
Full pod spec + events |
k8s_get_configmaps |
ConfigMaps with 5G field extraction |
k8s_get_nodes |
Node capacity, DPDK/SR-IOV labels |
k8s_get_events |
Cluster events |
5G Telecom-specific tools
| Tool | Description |
|---|---|
fiveg_core_topology |
Full NF map: pods, SBI endpoints, PLMN, slices |
fiveg_nf_status |
Deep status for a specific NF (AMF/SMF/UPF/…) |
fiveg_upf_dataplane |
N3/N4/N6/N9 interfaces, DPDK, hugepages |
fiveg_slice_info |
S-NSSAI / DNN / PLMN from ConfigMaps |
fiveg_health_check |
Full health report with recommendations |
MCP Resources (static reference)
5g://nf-reference– 3GPP TS 23.501 NF descriptions, SBI APIs, interface mapping5g://interface-map– N1–N26 reference-point descriptions
Supported 5G NFs (auto-detected from pod/deployment names & labels)
AMF · SMF · UPF · NRF · AUSF · UDM · UDR · PCF · NSSF · BSF · CHF · AF · N3IWF · SEPP
Tested against: Open5GS, free5GC, SD-Core, OAI-CN5G
Requirements
- Python ≥ 3.11
kubectlconfigured (in-cluster or local~/.kube/config)
Installation
pip install -r requirements.txt
Running the server (stdio)
python server.py
The server communicates over stdio using the MCP protocol.
It auto-detects in-cluster config, then falls back to ~/.kube/config.
If neither is available it runs in mock mode with representative 5G core data.
HTTP Streamable transport (network-accessible)
This project also supports the MCP Streamable HTTP transport, which exposes an HTTP/SSE endpoint suitable for remote clients and deployments in Kubernetes.
Run locally:
export MCP_TRANSPORT=http
export MCP_HTTP_HOST=0.0.0.0
export MCP_HTTP_PORT=8000
# Require auth (recommended)
export DANGEROUSLY_OMIT_AUTH=false
export MCP_HTTP_BEARER_TOKEN="<your-long-random-token>"
python server.py
# Health checks
curl http://localhost:8000/health
curl http://localhost:8000/ready
MCP endpoint URL: http://localhost:8000/mcp
Register with Claude (example):
claude mcp add --transport http k8s-5g http://localhost:8000/mcp
Kubernetes deployment is provided under k8s/. The container image defaults to
MCP_TRANSPORT=http and exposes port 8000 with readiness/liveness probes.
Notes:
- Streamable HTTP requires Accept headers that include both
application/jsonandtext/event-streamfor POST requests. - The server uses SSE for streaming responses by default.
Authentication (HTTP transport)
When running with MCP_TRANSPORT=http, this server supports a simple Bearer token guard on the /mcp endpoint.
- Set
DANGEROUSLY_OMIT_AUTH=false(default in Docker image) to enforce auth - Provide the secret in
MCP_HTTP_BEARER_TOKEN - Health endpoints (
/health,/ready) remain unauthenticated
Examples:
# Start server locally with a token
export MCP_TRANSPORT=http
export DANGEROUSLY_OMIT_AUTH=false
export MCP_HTTP_BEARER_TOKEN="s3cret-EXAMPLE-TOKEN"
python server.py
# Unauthorized request (401)
curl -i -X POST \
-H 'Accept: application/json, text/event-stream' \
-H 'Content-Type: application/json' \
--data '{"jsonrpc":"2.0","id":"1","method":"initialize","params":{}}' \
http://localhost:8000/mcp
# Authorized request (will advance handshake rather than 401)
curl -i -X POST \
-H 'Authorization: Bearer s3cret-EXAMPLE-TOKEN' \
-H 'Accept: application/json, text/event-stream' \
-H 'Content-Type: application/json' \
--data '{"jsonrpc":"2.0","id":"1","method":"initialize","params":{}}' \
http://localhost:8000/mcp
Kubernetes:
# Create secret with your token
kubectl -n mcp5g create secret generic mcp5g-auth \
--from-literal=token="s3cret-EXAMPLE-TOKEN"
# Apply namespace/RBAC/deployment
kubectl apply -f k8s/namespace.yaml
kubectl apply -f k8s/rbac.yaml
kubectl apply -f k8s/deployment.yaml
Security notes:
- Use a long, random token. Rotate via Secret update and rollout.
- For public endpoints, place this behind TLS and a network perimeter/proxy.
- For advanced OAuth 2.1 integration, consider wiring
mcp.server.authas a future enhancement.
Claude Desktop integration (stdio)
Edit ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"k8s-5g": {
"command": "python",
"args": ["/absolute/path/to/k8s-5g-mcp/server.py"],
"env": {
"KUBECONFIG": "/home/youruser/.kube/config"
}
}
}
}
Example LLM prompts
Show me the full 5G core topology in my cluster.
What is the health status of the AMF?
Are there any PFCP errors in the UPF logs?
What S-NSSAIs (slices) are configured?
Is DPDK enabled on my UPF node?
Run a full 5G core health check and give me recommendations.
Show me all pods in the open5gs namespace and classify them by NF type.
Architecture
LLM (Claude Desktop / any MCP client)
│ MCP (stdio / JSON-RPC 2.0)
▼
┌─────────────────────────────────────┐
│ server.py (MCP Server) │
│ ┌───────────┐ ┌─────────────────┐ │
│ │ k8s_client│ │telecom_analyzer │ │
│ │ (k8s SDK)│ │ (3GPP logic) │ │
│ └─────┬─────┘ └────────┬────────┘ │
└────────┼─────────────────┼──────────┘
│ kubernetes API │ ConfigMap / log parsing
▼
Kubernetes Cluster
(5G Core workloads: AMF, SMF, UPF, NRF, …)
Mock mode
If no kubeconfig is found, the server starts in mock mode and returns
a representative open5gs-style cluster with 8 NFs, DPDK on worker-2,
and pre-seeded log entries for testing.
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。