microsoft / microsoft/mcp

[EPIC] Agent-Assisted Azure Observability Workflows

Open
#3,114 0 comments 0 reactions 0 assignees View on GitHub
epic
Dominant language
C#
Stars
3.7k
Forks
624
Avg merge
2d 20h
Merged PRs (30d)
220

Description

## Problem statement

Azure MCP exposes individual monitoring, ingestion, diagnostics, and query capabilities, but users still need to assemble them manually into investigation workflows. Missing execution context, limited query results, disconnected ingestion tools, and isolated troubleshooting commands prevent agents from moving coherently from symptom to evidence and remediation.

## Vision

An agent can investigate an Azure operational problem end to end: identify the affected resource, collect relevant telemetry, validate or ingest data when needed, diagnose the failure, preserve execution context, and present an actionable remediation path.

## Current open work

- [ ] #2704, deliver disk performance troubleshooting
- [ ] #2694, add the `azmcp insights get` command
- [ ] #295, validate Azure Monitor ingestion data
- [ ] #288, upload Azure Monitor ingestion data
- [ ] #2714, send failing Application Insights operation context to Copilot in VS Code
- [ ] #1786, expose Kusto query execution statistics
- [ ] #280, remove AzureDiagnostics-only query limitations

## Completed foundation

- #671 added Application Insights trace listing
- #387 introduced Application Insights diagnostic commands
- #1801 established Azure Monitor instrumentation guidance

## Goals (in scope)

- Define representative diagnose-to-remediate user journeys
- Connect resource context, telemetry retrieval, diagnostics, and query evidence
- Preserve execution metadata needed for agents to explain findings reliably
- Provide safe ingestion and validation capabilities for supported scenarios
- Return structured outputs that support follow-up investigation and remediation

## Non-goals (out of scope)

- Building a replacement for Azure Monitor, Application Insights, or Azure Data Explorer
- Grouping unrelated monitoring commands without a validated user journey
- Automatically applying remediation without explicit authorization and safety controls

## Success criteria

- [ ] At least one end-to-end observability workflow is validated against a real Azure scenario
- [ ] Diagnostic results retain resource, operation, query, and timing context
- [ ] Monitoring and query tools expose the evidence required for agent reasoning
- [ ] Ingestion operations validate data and surface failures explicitly
- [ ] Workflow documentation and evaluation prompts cover supported investigation paths

## Dependencies

- Azure Monitor, Application Insights, and Azure Data Explorer APIs
- Structured command responses and reliable tool selection
- Agent-host support for preserving and presenting investigation context
- Security and authorization controls for telemetry and remediation operations

Contributor guide

Open the contributing guide

Research direction

Start by reviewing the linked open work items (#2704, #2694, #295, #288, #2714, #1786, and #280) alongside the stated dependencies and success criteria. Define a representative Azure investigation journey and identify the existing monitoring, ingestion, diagnostics, and query entry points it requires; done means one end-to-end workflow is validated against a real scenario with preserved context and documented evaluation coverage.

Written by the indexing model from the issue text.

Assessment

Tech stack
azure, csharp
Domain
cloud, observability-sre
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Quiet
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.