用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/stacklok/toolhive --skill deploying-vmcp-locally命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | deploying-vmcp-locally |
| description | Deploys a VirtualMCPServer configuration locally for manual testing and verification |
This skill helps you deploy and test VirtualMCPServer configurations in a local Kind cluster for manual verification.
Before using this skill, ensure you have:
If no Kind cluster exists, create one with the ToolHive operator:
# From the toolhive repository root
task kind-with-toolhive-operator
This creates a Kind cluster named toolhive with:
If you need to test local code changes:
# Set up cluster with e2e port mappings
task kind-setup-e2e
# Install CRDs
task operator-install-crds
# Build and deploy local operator image
task operator-deploy-local
Apply the YAML configuration you want to test:
kubectl apply -f <path-to-vmcp-yaml> --kubeconfig kconfig.yaml
Check the VirtualMCPServer status:
# List all VirtualMCPServers
kubectl get virtualmcpserver --kubeconfig kconfig.yaml
# Get detailed status
kubectl get virtualmcpserver <name> -o yaml --kubeconfig kconfig.yaml
# Check operator logs for issues
kubectl logs -n toolhive-system -l app.kubernetes.io/name=thv-operator --kubeconfig kconfig.yaml
For NodePort service type (useful for local testing):
# Get the NodePort
kubectl get svc vmcp-<name> -o jsonpath='{.spec.ports[0].nodePort}' --kubeconfig kconfig.yaml
# Test the endpoint (port will be on localhost when using kind-setup-e2e)
curl http://localhost:<nodeport>/mcp
For ClusterIP (default), use port-forward:
kubectl port-forward svc/vmcp-<name> 4483:4483 --kubeconfig kconfig.yaml
curl http://localhost:4483/mcp
Use an MCP client to verify tool discovery and execution:
# Initialize MCP session
curl -X POST http://localhost:<port>/mcp \
-H "Content-Type: application/json" \
-d '{"jsonrpc": "2.0", "method": "initialize", "params": {"protocolVersion": "2024-11-05", "capabilities": {}, "clientInfo": {"name": "test", "version": "1.0"}}, "id": 1}'
# List tools
curl -X POST http://localhost:<port>/mcp \
-H "Content-Type: application/json" \
-d '{"jsonrpc": "2.0", "method": "tools/list", "id": 2}'
When done testing:
# Remove specific resources
kubectl delete -f <path-to-vmcp-yaml> --kubeconfig kconfig.yaml
# Or destroy the entire cluster
task kind-destroy
Reference example configurations are in examples/operator/virtual-mcps/:
| File | Description |
|---|---|
vmcp_simple_discovered.yaml | Basic discovered mode configuration |
vmcp_conflict_resolution.yaml | Tool conflict handling strategies |
vmcp_inline_incoming_auth.yaml | Inline authentication configuration |
vmcp_production_full.yaml | Full production configuration |
composite_tool_simple.yaml | Simple composite tool workflow |
composite_tool_complex.yaml | Complex multi-step workflows |
composite_tool_with_elicitations.yaml | Workflows with user prompts |
Check that:
kubectl describe virtualmcpserver <name> --kubeconfig kconfig.yaml
kubectl get mcpgroup --kubeconfig kconfig.yaml
kubectl get mcpserver --kubeconfig kconfig.yaml
Verify backend servers have the correct groupRef:
kubectl get mcpserver -o custom-columns=NAME:.metadata.name,GROUP:.spec.groupRef --kubeconfig kconfig.yaml
For testing, use anonymous auth:
incomingAuth:
type: anonymous
authzConfig:
type: inline
inline:
policies:
- 'permit(principal, action, resource);'