Azure Omni-Tool MCP Server
# Azure Omni-Tool MCP Server
A **Model Context Protocol (MCP) server** in TypeScript that acts as an intelligent bridge between natural language requests and Azure CLI execution.
## Features
✅ **Plan/Execute Flow** - Review commands before execution
✅ **Safety Guardrails** - Shell injection detection, destructive command warnings
✅ **Audit Trail** - Operator email tagging for traceability
✅ **Retry Logic** - Exponential backoff for transient failures
✅ **Caching** - LRU cache with configurable TTL
✅ **Tenant Scoping** - Configure tenant/subscription via environment
✅ **Azure Service Adapters** - Type-safe access to 8 Azure services
---
## Architecture Overview
```mermaid
flowchart TB
subgraph Client["🖥️ Client Layer"]
LLM[LLM / AI Agent]
end
subgraph MCP["⚙️ MCP Server"]
direction TB
Entry[index.ts]
subgraph Tools["Tools"]
T1[manage_azure_resources]
T2[get_azure_context]
T3[azure_service]
end
subgraph Lib["Core Libraries"]
Auth[auth.ts]
Cache[cache.ts]
CLI[cli-executor.ts]
Retry[retry.ts]
Safety[safety.ts]
Audit[audit.ts]
end
subgraph Services["Service Adapters"]
S1[StorageService]
S2[CosmosService]
S3[SearchService]
S4[KustoService]
S5[MonitorService]
S6[AppConfigService]
S7[KeyVaultService]
S8[PostgresService]
end
end
subgraph Azure["☁️ Azure"]
AzCLI[Azure CLI]
AzAPI[Azure APIs]
end
LLM -->|MCP Protocol| Entry
Entry --> Tools
Tools --> Lib
Tools --> Services
Services --> Lib
Lib --> AzCLI
Auth --> AzAPI
```
---
## Request Flow
```mermaid
sequenceDiagram
participant C as Client
participant M as MCP Server
participant S as Safety
participant E as CLI Executor
participant A as Azure
C->>M: Tool Request
M->>S: Validate Input
alt Unsafe Command
S-->>M: Block + Warning
M-->>C: Error Response
else Safe
S-->>M: Approved
M->>E: Execute Command
E->>A: az CLI call
A-->>E: Response
E-->>M: Result + Parse
M-->>C: Structured Output
end
```
---
## Plan/Execute Flow
```mermaid
flowchart LR
A[LLM Client] -->|Natural Language| B[MCP Server]
B --> C{execute_now?}
C -->|false| D[Return Plan]
C -->|true| E[Execute CLI]
E --> F{Success?}
F -->|Yes| G[Return Output]
F -->|No| H[Return Error + Analysis]
H -->|Feedback Loop| A
```
---
## Quick Start
### 1. Install Dependencies
```bash
npm install
```
### 2. Configure Environment
```bash
cp .env.example .env
# Edit .env with your settings
```
### 3. Build & Run
```bash
npm run build
npm start
```
### MCP Client Configuration
```json
{
"mcpServers": {
"azure-omni-tool": {
"command": "node",
"args": ["path/to/Azure-mcp/dist/index.js"]
}
}
}
```
---
## Tools
### `manage_azure_resources`
Plan and execute Azure CLI commands with safety checks.
| Argument | Type | Description |
|----------|------|-------------|
| `command` | string | Azure CLI command |
| `explanation` | string | Why this command was chosen |
| `execute_now` | boolean | `false` = plan, `true` = execute |
### `get_azure_context`
Query Azure environment with caching.
| Query Type | Description |
|------------|-------------|
| `subscriptions` | List accessible subscriptions |
| `resource_groups` | List resource groups |
| `resources` | List resources |
| `custom` | Custom KQL via Resource Graph |
### `azure_service`
Interact with specific Azure services.
| Service | Actions |
|---------|---------|
| `storage` | list, listContainers, listBlobs, getContainer, listTables, queryTable |
| `cosmos` | list, listDatabases, listContainers, query, getContainer |
| `search` | list, listIndexes, getIndex, query, getService |
| `kusto` | list, listDatabases, listTables, getSchema, sample, query |
| `monitor` | list, getWorkspace, listTables, query, listMetrics, getMetrics |
| `appconfig` | list, getStore, listKeyValues, getKeyValue, setKeyValue, lock, unlock |
| `keyvault` | list, getVault, listKeys, getKey, createKey, listSecrets, getSecret, listCertificates |
| `postgres` | list, getServer, listDatabases, listParameters, getParameter, listTables, getTableSchema, query |
---
## Environment Variables
| Variable | Description | Default |
|----------|-------------|---------|
| `AZURE_TENANT_ID` | Azure tenant for scoping | - |
| `AZURE_SUBSCRIPTION_ID` | Default subscription | - |
| `OPERATOR_EMAIL` | Email for audit trail | - |
| `OPERATOR_NAME` | Operator name | - |
| `LOG_LEVEL` | Logging level | `info` |
| `ENABLE_CACHE` | Enable query caching | `true` |
| `CACHE_TTL_SECONDS` | Cache duration | `300` |
| `CACHE_CLEANUP_INTERVAL_MS` | Cache cleanup interval | `60000` |
| `MAX_RETRIES` | Retry attempts | `3` |
| `RETRY_DELAY_MS` | Base retry delay | `1000` |
| `COMMAND_TIMEOUT_MS` | CLI timeout | `120000` |
| `AZURE_MCP_INCLUDE_PRODUCTION_CREDENTIALS` | Enable Managed Identity | `false` |
---
## Project Structure
```
Azure-mcp/
├── src/
│ ├── index.ts # MCP server entry
│ ├── lib/
│ │ ├── auth.ts # Azure credential management
│ │ ├── audit.ts # Audit trail with correlation IDs
│ │ ├── cache.ts # LRU cache with TTL
│ │ ├── cli-executor.ts # Azure CLI wrapper
│ │ ├── config.ts # Environment config
│ │ ├── logger.ts # Structured JSON logging
│ │ ├── retry.ts # Exponential backoff
│ │ ├── safety.ts # Input sanitization
│ │ └── types.ts # Shared types
│ ├── services/
│ │ ├── base-service.ts # Abstract service base
│ │ ├── storage.ts # Azure Storage
│ │ ├── cosmos.ts # Cosmos DB
│ │ ├── search.ts # AI Search
│ │ ├── kusto.ts # Data Explorer
│ │ ├── monitor.ts # Monitor / Log Analytics
│ │ ├── appconfig.ts # App Configuration
│ │ ├── keyvault.ts # Key Vault
│ │ ├── postgres.ts # PostgreSQL Flexible Server
│ │ └── index.ts # Service factory
│ └── tools/
│ ├── azure-manager.ts # Plan/Execute tool
│ ├── context-retriever.ts # Context queries
│ └── service-tool.ts # Service adapter tool
├── .env.example
├── package.json
└── tsconfig.json
```
---
## Prerequisites
- Node.js >= 18.0.0
- Azure CLI installed and authenticated (`az login`)
---
## License
MIT
TDQS
Scored across 3 tools
The three tools have overlapping and unclear boundaries. 'azure_service' appears to handle data querying across multiple Azure services, while 'manage_azure_resources' seems to cover CLI-based operations, but their scopes are not clearly differentiated. 'get_azure_context' is distinct for context retrieval, but the other two tools could easily be confused for similar purposes, leading to agent misselection.
Naming is inconsistent across the tool set. 'azure_service' uses snake_case, 'get_azure_context' follows a verb_noun pattern with snake_case, and 'manage_azure_resources' uses a verb_noun pattern but with a different verb style ('manage' vs. 'get'). This mixed convention reduces predictability and readability for agents.
With 3 tools, the count is borderline for a server named 'Azure Omni-Tool MCP Server', which suggests broad Azure coverage. This feels thin given the extensive Azure domain, as many operations might be crammed into a few tools, but it's not critically low. A more appropriate scope might include more specialized tools for better granularity.
The tool surface has significant gaps for an Azure-focused server. While 'azure_service' covers querying across services and 'manage_azure_resources' handles CLI operations, there are missing core operations like creating, updating, or deleting resources directly. The server lacks clear CRUD lifecycle coverage, which could cause agent failures in common Azure workflows.