🤖 Lab 3: Using Dynatrace MCP for Agentic AI

Duration: ~30 minutes

In this lab, you’ll configure and use the Dynatrace MCP (Model Context Protocol) server to interact with Dynatrace directly from your IDE using AI assistants like GitHub Copilot.

🎯 Learning Objectives

• Understand what MCP is and how it enables agentic AI
• Install and configure the Dynatrace MCP server
• Use natural language to query Dynatrace from VS Code
• Analyze problems and traces using AI assistance
• Explore real-world use cases for observability-driven AI

What is MCP?

Model Context Protocol (MCP) is an open standard that allows AI assistants to interact with external tools and data sources. With Dynatrace MCP, you can:

• Query Dynatrace using natural language
• Analyze problems and incidents
• Retrieve metrics, traces, and logs
• Get AI-powered insights about your applications

Think of it as giving your AI assistant direct access to Dynatrace!

Step 1: Configure the Dynatrace MCP Server

The Dynatrace MCP server is already pre-configured in this workshop! The authentication token was automatically configured when you ran the setup script in Lab 0.

Step 2: Verify MCP Connection

2.1 Open GitHub Copilot Chat

1. Click on the Copilot icon on the top bar
2. Or use keyboard shortcut: Cmd+Shift+I (Mac) / Ctrl+Shift+I (Windows)

2.2 Test the Connection

In the Copilot chat, type:

@dynatrace What services are available in my environment?

If configured correctly, Copilot will query Dynatrace and return a list of services!

🎭 New! Dynatrace Intelligence ✨

Dynatrace Intelligence provides agentic workflows that go beyond simple queries. These exercises showcase the latest capabilities for each persona. What this means is that with Dynatrace’s MCP server, your personal AI agent (GitHub CoPilot for this lab) can actually interact with various agents provided by Dynatrace! When running the below queries, notice how your AI agent will interact with various agents from Dynatrace (e.g. Grail Query Agent, Data Analysis Agent, Root Cause Agent, etc.).

Learn more here:

• Dynatrace Intelligence - Announcement
• Dynatrace Intelligence - Documentation

🎭 Your Mission (Choose Your Persona)

From this point forward, you’ll focus on different use cases depending on your role.

@dynatrace Tell me about the service called ai-chat-service-{YOUR_ATTENDEE_ID}

@dynatrace What is the total input and output token usage for spans in regards to the ai-chat-service-{YOUR_ATTENDEE_ID}

I'm looking at main.py where I call Azure OpenAI.
@dynatrace What's the average latency for Azure OpenAI calls from my ai-chat-service-{YOUR_ATTENDEE_ID} service?

I'm seeing slow responses in my RAG pipeline. 
@dynatrace Analyze the trace data and tell me which step is the bottleneck for my ai-chat-service-{YOUR_ATTENDEE_ID} service.
Is it embeddings, vector search, or the LLM call?

@dynatrace Analyze my ai-chat-service-{YOUR_ATTENDEE_ID} service and suggest optimizations to reduce token usage while maintaining response quality

@dynatrace Help me understand why some of my RAG queries might be slow. Look at the trace data for patterns.

@dynatrace Show me all errors in the last 15 minutes for my ai-chat-service-{YOUR_ATTENDEE_ID} service.
What error codes are appearing most frequently?

@dynatrace Show me the log entries for EMB_NULL_VECTOR errors in my ai-chat-service-{YOUR_ATTENDEE_ID}.
Include the error_message, stage, and timestamp for each.

@dynatrace For the errors in my ai-chat-service-{YOUR_ATTENDEE_ID} service, 
show me all the log attributes like error_code, error_message, and stage.
What patterns do you see?

@dynatrace I'm seeing CTX_WINDOW_EXCEEDED errors in my ai-chat-service-{YOUR_ATTENDEE_ID} service. 
Show me the full log details for these errors and explain what's happening.

@dynatrace What's causing the errors in my ai-chat-service-{YOUR_ATTENDEE_ID} service?
Analyze the error logs and tell me which component is failing most often.

@dynatrace Generate a DQL query to find all logs where error_code equals 'EMB_NULL_VECTOR' 
for my ai-chat-service-{YOUR_ATTENDEE_ID} service in the last hour.

@dynatrace Tell me about the service called ai-chat-service-{YOUR_ATTENDEE_ID}

@dynatrace Are there any anomalies in the last hour for my ai-chat-service-{YOUR_ATTENDEE_ID} service?
What's the current error rate and how does it compare to the baseline?

@dynatrace What is the total input and output token usage for spans in regards to the ai-chat-service-{YOUR_ATTENDEE_ID} service?

@dynatrace Are there any open problems affecting my ai-chat-service-{YOUR_ATTENDEE_ID} service?
If so, what's the root cause and which services are impacted?
Draft a Slack message summarizing the incident.

@dynatrace Generate a summary of my ai-chat-service-{YOUR_ATTENDEE_ID} service's architecture based on the service flow data

@dynatrace Analyze my ai-chat-service-{YOUR_ATTENDEE_ID} service and suggest optimizations to reduce token usage while maintaining response quality

@dynatrace Give me a quick summary of all errors hitting my ai-chat-service-{YOUR_ATTENDEE_ID} service in the last 15 minutes.
How many errors occurred? What types? What's the error rate percentage?

@dynatrace Show me the error log entries for my ai-chat-service-{YOUR_ATTENDEE_ID} service.
Include the error_code, error_message, stage, and timestamp for each error.

@dynatrace When did errors start occurring in my ai-chat-service-{YOUR_ATTENDEE_ID} service?
Show me the timeline of errors over the last 15 minutes.

@dynatrace What percentage of requests to my ai-chat-service-{YOUR_ATTENDEE_ID} service are failing?
Is this affecting all users or just specific request types?

@dynatrace What are the most common error_code values in my ai-chat-service-{YOUR_ATTENDEE_ID} service?
Rank them by frequency.

@dynatrace Analyze the error patterns in my ai-chat-service-{YOUR_ATTENDEE_ID} service.
What is Dynatrace Intelligence's assessment of the root cause?
Which component is the source of the failures?

@dynatrace I need to communicate an incident to stakeholders.
Based on the errors in my ai-chat-service-{YOUR_ATTENDEE_ID} service, draft a brief incident summary
including: affected service, error types, error rate, and preliminary root cause.

Bonus - Dynatrace Intelligence in the Dynatrace Platform

Want to make use of Dynatrace Intelligence before even setting up the MCP? Try it out directly in the Dynatrace UI.

1. Open a Dynatrace Notebook
2. Add a ✨ Prompt tile
3. Try out the following question:

   I would like to see top 5 spans summarized by token usage and span name for service "ai-chat-service-{YOUR_ATTENDEE_ID}" in descending order
   

Pro tip: This is just one example. Dynatrace Intelligence is available in several parts of the platform including the Problems app, Notebooks and Dashboards, Logs app, Databases app, and more!

Learn more here:

• Dynatrace Intelligence - Announcement
• Dynatrace Intelligence - Documentation

🎭 Going Beyond the Basics

Below are some more advanced examples to get the most out of Dynatrace Intelligence and your AI agents!

@dynatrace Look at the error traces for my ai-chat-service-{YOUR_ATTENDEE_ID} service.
Can you identify which function or code path is causing the failures?
Provide code-level details and stack traces if available.

@dynatrace Based on the EMB_NULL_VECTOR errors in my ai-chat-service-{YOUR_ATTENDEE_ID} service, 
what code changes would you recommend to handle this error gracefully?
Show me a Python code example for proper error handling.

@dynatrace Have there been any recent deployments or configuration changes 
to my ai-chat-service-{YOUR_ATTENDEE_ID} service that correlate with the increase in errors?

@dynatrace Generate a DQL query to find all logs where error_code equals 'CTX_WINDOW_EXCEEDED' 
for my ai-chat-service-{YOUR_ATTENDEE_ID} service in the last hour.
Explain what the query does.

@dynatrace For the errors in my ai-chat-service-{YOUR_ATTENDEE_ID} service, 
show me the full dependency map. Which downstream services are affected?
What's the blast radius of this issue?

@dynatrace Create a comprehensive incident report for the issues affecting 
my ai-chat-service-{YOUR_ATTENDEE_ID} service. Include:
- Timeline of when errors started
- Affected components and their relationships
- Error counts by type
- Recommended severity level

@dynatrace Based on the current error patterns in my ai-chat-service-{YOUR_ATTENDEE_ID} service,
what remediation actions would you recommend?
Are there any runbooks or automation workflows that could help resolve this?

@dynatrace Generate a DQL query for a dashboard tile that shows:
- Error rate over time for my ai-chat-service-{YOUR_ATTENDEE_ID} service
- Breakdown by error_code
- 5-minute time buckets for the last hour

Step 7: MCP Best Practices

7.1 Effective Prompting

Good prompts are specific:

✅ Good: @dynatrace Show me the P95 response time for my ai-chat-service-{YOUR_ATTENDEE_ID} service over the last 4 hours

❌ Vague: @dynatrace How is my service doing?

7.2 Iterative Queries

Start broad, then drill down:

1. @dynatrace Show me an overview of my AI service
2. @dynatrace What are the slowest endpoints?
3. @dynatrace Why is /chat endpoint slow?
4. @dynatrace Show me slow traces for /chat

7.3 Combining with Code

You can use MCP alongside your code:

I'm looking at main.py where I call Azure OpenAI. 
@dynatrace What's the average latency for Azure OpenAI calls from my service?

✅ Checkpoint

Before completing the workshop, verify:

• You’ve added your DT_ENVIRONMENT URL to .vscode/mcp.json
• You’ve reloaded VS Code after saving the configuration
• You can query Dynatrace using @dynatrace in Copilot Chat
• You’ve successfully retrieved information about your AI service
• You’ve used the 🐛 Simulate Errors toggle to generate test errors
• You’ve investigated errors using MCP without leaving your IDE
• You understand how to use MCP for problem analysis
• You’ve explored the Dynatrace Intelligence agentic workflow exercises

🆘 Troubleshooting

“MCP server not found”

1. Verify Node.js is installed: node --version
2. Ensure the .vscode/mcp.json file exists in your workspace
3. Test MCP server access: npx @dynatrace-oss/dynatrace-mcp-server@latest --version
4. Clear NPM cache if needed: npm cache clean --force

“@dynatrace not recognized”

1. Check that .vscode/mcp.json contains the correct configuration
2. Verify the JSON syntax is valid (no trailing commas, proper quotes)
3. Reload VS Code window (Developer: Reload Window)
4. Make sure DT_ENVIRONMENT is set to your Dynatrace URL

“Authentication failed” or “401 Unauthorized”

1. Verify DT_ENVIRONMENT in .vscode/mcp.json is correct
2. Ensure the URL format is https://YOUR_ENV_ID.apps.dynatrace.com
3. Check that your Dynatrace environment allows API access
4. Check that your token has appropriate permissions:
   • Read entities
   • Read problems
   • Read metrics
   • Read logs
   • Read traces

“No data returned”

1. Verify DT_ENVIRONMENT in .vscode/mcp.json is correct
2. Check that your service is sending data to Dynatrace
3. Try a simpler query first: @dynatrace List all services

“Connection refused” or “Network error”

1. If in a Codespace, ensure outbound connections are allowed
2. Check if your organization has firewall rules blocking the connection
3. Verify the URL format is https://YOUR_ENV_ID.apps.dynatrace.com

🎓 What You’ve Learned

🚀 Next Steps

Now that you’ve completed this lab, continue to Lab 4 to learn how to automate your AI observability workflows!

🎉 Great Progress!

You’ve learned how to use Dynatrace MCP for agentic AI workflows. Now let’s put it all together with automated workflows!