Architecture AI/MCP

flowchart TB subgraph CLIENTS["External Clients"] direction LR OC["🖥️ OpenShift Console\nLightspeed Plugin"] N8N_EXT["⚙️ n8n :443\nWorkflows"] MCPI["🔍 MCP Inspector"] CURL["📡 curl / SDK"] end subgraph OLS_NS["Namespace: openshift-lightspeed"] OLS["🤖 lightspeed-app-server\nOLS v1.0.10\nfeatureGate: MCP"] PROXY["🔄 tool-choice-proxy :4001\n─────────────────\n• model remap llama→deepseek\n• tool_choice injection\n• SSE bridge\n• budget fail → fallback qwen-cpu"] LITELLM["📊 LiteLLM :4000\n─────────────────\nmodels: deepseek, llama-scout,\nqwen-cpu │ PostgreSQL"] N8N_INT["📋 n8n :5678\n17 workflows + Mailpit"] subgraph MCP_SERVERS["MCP Servers"] direction LR OCP_MCP["🔧 openshift-mcp-server :8080\nQuarkus · 19 tools\nocp_* prefix"] K8S_MCP["🔧 kubernetes-mcp-server :8085\nGo · 19 tools\nk8s_* prefix"] end end subgraph MCP_NS["Namespace: mcp-system"] GW["🌐 MCP Gateway\nKuadrant / Envoy\n─────────────────\n38 federated tools on /mcp\nocp_* → openshift-mcp-server\nk8s_* → kubernetes-mcp-server"] end subgraph MODELS["LLM Models"] direction LR MAAS["☁️ MaaS LiteLLM\ndeepseek-r1-distill-qwen-14b\nEXTERNAL"] VLLM["🖥️ vLLM CPU · KServe\nQwen2.5-7B-Instruct :8080\nRHOAI + Serverless\n~170s/response"] end K8S_API["⚡ Kubernetes API Server"] OC -->|"question"| OLS OLS -->|"OpenAI API"| PROXY OLS -->|"MCP Protocol\n(direct)"| OCP_MCP OLS -->|"MCP Protocol\n(direct)"| K8S_MCP PROXY --> LITELLM N8N_EXT --> N8N_INT N8N_INT --> LITELLM LITELLM -->|"budget ✅"| MAAS LITELLM -->|"budget ❌\nfallback"| VLLM MCPI --> GW CURL --> GW GW --> OCP_MCP GW --> K8S_MCP OCP_MCP -->|"cluster-admin"| K8S_API K8S_MCP -->|"cluster-admin"| K8S_API

sequenceDiagram actor User as 👤 User participant Console as OpenShift Console
Lightspeed Plugin participant OLS as OLS v1.0.10
lightspeed-app-server participant Proxy as tool-choice-proxy
:4001 participant LiteLLM as LiteLLM
:4000 participant MaaS as MaaS
deepseek-r1 participant vLLM as vLLM CPU
qwen-2.5-7b participant MCP as MCP Server
openshift / kubernetes User->>Console: Ask question Console->>OLS: Forward query OLS->>MCP: Discover tools (MCP protocol) MCP-->>OLS: Available tools list OLS->>Proxy: chat/completions + tools Note over Proxy: Remap model:
llama-scout → deepseek-r1
Inject tool_choice
Set stream=false Proxy->>LiteLLM: Modified request LiteLLM->>MaaS: Forward to model alt Budget OK MaaS-->>LiteLLM: tool_call response else Budget Exceeded MaaS-->>LiteLLM: 400 budget error LiteLLM-->>Proxy: Error response Note over Proxy: Detect budget error
Retry with fallback model Proxy->>LiteLLM: Retry → qwen-2.5-7b-cpu LiteLLM->>vLLM: Forward to CPU model vLLM-->>LiteLLM: tool_call response (~170s) end LiteLLM-->>Proxy: tool_call JSON Note over Proxy: Convert to SSE stream
for OLS compatibility Proxy-->>OLS: SSE stream with tool_call OLS->>MCP: Execute tool (e.g. checkClusterHealth) MCP-->>OLS: Tool result OLS->>Proxy: chat/completions + tool result Proxy->>LiteLLM: Forward LiteLLM-->>Proxy: Final text response Proxy-->>OLS: SSE stream OLS-->>Console: Formatted answer Console-->>User: Display response

flowchart LR subgraph CORE["Core Components"] direction TB A["🔧 openshift-mcp-server\nQuarkus · 19 tools · ocp_*"] B["🔧 kubernetes-mcp-server\nGo · 19 tools · k8s_*"] end subgraph PROXY_LAYER["Proxy & Model Layer"] direction TB C["🔄 tool-choice-proxy\nStreaming fix · tool_choice\nModel remap · CPU fallback"] D["📊 LiteLLM\nMulti-model proxy\nPostgreSQL · API keys"] end subgraph MODELS_REF["Models"] direction TB E["☁️ MaaS deepseek-r1\nPrimary · External"] F["🖥️ vLLM qwen-2.5-7b\nFallback · Local CPU"] end subgraph EXTERNAL["External Access"] direction TB G["🌐 MCP Gateway\nKuadrant/Envoy\n38 federated tools"] end A --> PROXY_LAYER B --> PROXY_LAYER C --> D D --> E D --> F G --> CORE

pie title 38 MCP Tools Distribution "k8s Read (13)" : 13 "k8s Write (6)" : 6 "ocp Observability (9)" : 9 "ocp Provisioning (5)" : 5 "ocp Benchmarks (5)" : 5

flowchart LR A["🔘 Manual\nTrigger"] --> B["⚙️ Set\nParameters"] B --> C["🔧 MCP\nTool Call"] C --> D["🤖 AI Analysis\nvia LiteLLM"] D --> E["📧 Email Report\nvia Mailpit"] style A fill:#2d2d2d,stroke:#CC0000,color:#e0e0e0 style B fill:#2d2d2d,stroke:#CC0000,color:#e0e0e0 style C fill:#2d2d2d,stroke:#CC0000,color:#e0e0e0 style D fill:#2d2d2d,stroke:#CC0000,color:#e0e0e0 style E fill:#2d2d2d,stroke:#CC0000,color:#e0e0e0

# 1. Initialize session
SESSION=$(curl -sk -X POST \
  https://mcp-gateway.apps.cluster-6tx4w.6tx4w.sandbox5380.opentlc.com/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"1.0"}}}' \
  -D - 2>/dev/null | grep -i mcp-session-id | tr -d '\r' | awk -F': ' '{print $2}')

# 2. Send initialized notification
curl -sk -X POST \
  https://mcp-gateway.apps.cluster-6tx4w.6tx4w.sandbox5380.opentlc.com/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "Mcp-Session-Id: $SESSION" \
  -d '{"jsonrpc":"2.0","method":"notifications/initialized"}'

# 3. List all 38 federated tools
curl -sk -X POST \
  https://mcp-gateway.apps.cluster-6tx4w.6tx4w.sandbox5380.opentlc.com/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "Mcp-Session-Id: $SESSION" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/list"}'

# 4. Call a tool (e.g. list pods in openshift-lightspeed)
curl -sk -X POST \
  https://mcp-gateway.apps.cluster-6tx4w.6tx4w.sandbox5380.opentlc.com/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "Mcp-Session-Id: $SESSION" \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"k8s_pods_list_in_namespace","arguments":{"namespace":"openshift-lightspeed"}}}'