Tools
Tools are the atomic capabilities your agent uses to take action — querying logs, running commands, executing code, searching documents, and sending notifications. Your agent selects the right tools automatically based on the task at hand.
| Category | What it covers | Setup |
|---|---|---|
| Built-in | Azure operations, diagnostics, monitoring, log queries, and visualization | None — available through managed identity |
| MCP | Any external service via Model Context Protocol | Add an MCP connector |
| Code execution | Python and shell execution in sandboxed environments | None — built-in |
| Knowledge | Document search, agent memory, application topology | None — built-in (some features require connectors) |
| Communication | Email and Teams notifications | Add Outlook or Teams connector |
| Incident management & DevOps | Incident platforms and source code repositories | Add platform connector |
| Custom tools | Your own Kusto, Python, Link, and HTTP tools | Create in Builder UI |
Tools combine with skills and custom agents to create powerful automation. Skills attach tools to procedural instructions. Custom agents get dedicated tool sets for their domain.
Built-in tools
Your agent includes tools for Azure operations, diagnostics, monitoring, and log queries. These work immediately through the agent's managed identity — no connector setup required. Ensure your agent has appropriate RBAC permissions on target resources.
Built-in tools cover the full operational spectrum: run Azure CLI commands, query Application Insights and Log Analytics, analyze Azure Monitor metrics, manage AKS clusters with kubectl, diagnose Container Apps, Function Apps, App Service, and more. Specialized diagnostic tools perform deeper analysis — CPU profiling, API Management diagnostics, deployment verification, reliability assessment, and remediation actions. Visualization tools generate charts and integrate with Grafana dashboards.
Your agent selects the right tool based on the resource type and the nature of your question. For a deeper look at Azure diagnostic capabilities, see Azure observability and Root cause analysis.
MCP tools
The Model Context Protocol (MCP) extends your agent with tools from any MCP-compatible server — Datadog, Splunk, GitHub, and more. When you connect an MCP server, your agent discovers its tools automatically and makes them available for custom agent assignment. Two transport types are supported: Streamable-HTTP for remote services and stdio for local processes.
For architecture details, transport types, partner connectors, health monitoring, and tool management patterns, see MCP Connectors & Tools.
Code execution
Your agent can write and execute code in sandboxed environments for data analysis, custom calculations, and report generation. The built-in Code Interpreter runs Python and shell commands in an isolated container — useful for processing query results, generating charts, and creating PDF reports.
You can also create reusable custom Python tools with your own pre-written functions and pip dependencies. Unlike the Code Interpreter (which generates code on the fly), custom Python tools run your defined logic with specific inputs.
See Python code execution for details.
Tool selection intelligence
Each tool includes a description prompt — a detailed instruction that the model reads when deciding which tool to use and how to use it. These prompts shape how your agent reasons about tool selection and execution:
- Parallel execution — When your agent identifies independent operations (such as multiple diagnostic commands that do not depend on each other), it runs them simultaneously in a single turn. Terminal commands, file searches, and other workspace operations all support parallel execution.
- Task delegation — For complex searches that would require multiple rounds of file pattern matching, content searching, and reading, the agent delegates to a specialized built-in Explore task agent via the Task tool. The task agent handles the multi-step search autonomously and returns a structured result.
- Tool routing — The agent selects the most appropriate tool for each operation. Simple file pattern searches use FileSearch directly, while complex exploration tasks use the Task tool. Shell commands run via RunInTerminal, while Python data analysis runs in the isolated code interpreter.
These tool prompts are continuously refined to improve reasoning quality and investigation speed.
Knowledge
Your agent uses knowledge tools to access organizational context and build understanding of your environment over time. Document search finds relevant procedures and runbooks from your knowledge base. Agent memory provides vector search across uploaded files. Application topology maps resource relationships and network connections.
Troubleshooting guide (TSG) retrieval finds and follows guides indexed from Azure DevOps wikis. The knowledge graph builds a persistent entity-relation model of your environment as your agent learns from investigations.
For more on how knowledge works, see Memory & Knowledge. To add documents, see Upload knowledge documents.
Communication
Send investigation findings through the channels your team uses. Connect Outlook to email summaries and reports with attachments. Connect Teams to post updates and reply to conversations in your channels.
Both require their respective connectors to be configured. See Send notifications for setup and usage.
Incident management and DevOps
Your agent also integrates with incident management platforms and source code repositories. These have dedicated concept pages:
-
Incident platforms — Connect PagerDuty or ServiceNow to receive alerts and manage incident lifecycles. See Incident platforms and Incident response.
-
DevOps — Connect GitHub or Azure DevOps to access repositories, pull requests, issues, and work items. See Set up Azure DevOps connector.
Custom tools
Create your own tools for operations specific to your environment. Four types are available:
| Type | Use case |
|---|---|
| Kusto | Run predefined KQL queries with parameter substitution |
| Python | Execute custom Python functions with pip dependencies |
| Link | Generate URLs from templates with dynamic parameters |
| HTTP client | Call REST APIs with authentication |
Custom tools are created through the Builder UI and can be attached to skills or assigned to custom agents. See Kusto tools and Python code execution.
Related
| Resource | Why it matters |
|---|---|
| Connectors | Connect external systems and activate connector-based tools |
| Skills | Combine tools with procedural instructions |
| Custom Agents | Specialists with dedicated tool sets |
| Permissions | RBAC configuration for Azure tool access |
| Memory & Knowledge | How knowledge tools build context over time |
| Incident platforms | Connect PagerDuty, ServiceNow, or other incident platforms |