Best practices for MCP prompting

Use precise terminology and avoid ambiguous language. Specify exactly what you want to achieve and include relevant context and constraints.

Good examples:

• "Show me the performance metrics for the latest run of the Open-telem-demo test"

• "Execute the API load test in the MCP-Demo workspace"

Avoid vague requests:

• "Check the test"

• "Run something"

Include relevant information to help the AI understand your context, such as:

• Workspace names or IDs

• Test names and their purposes

• Time frames for analysis

• Expected outcomes or thresholds

Good examples:

• "Why did the overnight regression test fail? I'm looking at result ID abc-123 from the QA workspace"

• "Compare the response times between yesterday's baseline test and today's feature branch test for the checkout API"

Start broad, then narrow down based on results. This conversation flow helps you discover and focus on what matters most:

1. "Show me recent test results" → Get overview

2. "Focus on the failed tests from last week" → Narrow scope

3. "What caused the failures in the payment service tests?" → Specific analysis

4. "Show me the error patterns and recommend fixes" → Actionable insights

Break complex tasks into clear, logical steps rather than overwhelming the AI with everything at once.

Instead of:

"I need to know everything about performance issues and create a report with recommendations and comparison data"

Use:

"Please help me create a performance analysis report. I need:
1. Performance metrics for test result XYZ
2. Comparison with the previous baseline
3. Identification of any bottlenecks
4. Recommendations for optimization
5. Format as an executive summary"

Structure: [Question] + [Context] + [Desired Action]

Example:

"What's causing the high response times [Question]
in our e-commerce API tests from the last 3 runs [Context]?
Please analyze the patterns and suggest specific optimizations [Action]."

Structure: Compare [Item A] with [Item B] focusing on [Specific Metrics]

Example:

"Compare the throughput and error rates between:
- Test result abc-123 (baseline)
- Test result def-456 (new feature)
Focus on the checkout and payment transactions."

Structure: [Problem Statement] + [Context] + [Investigation Request]

Example:

"Our API response times increased by 40% after the last deployment.
Looking at workspace 'Production-Tests', test 'API-Regression'.
Can you analyze the recent results and identify what changed?"

Structure: Create [Report Type] for [Audience] including [Specific Elements]

Example:

"Create an executive performance summary for the development team including:
- Key metrics from this week's tests
- Trend analysis vs. last month
- Top 3 performance concerns
- Recommended actions with priorities"

1. Discovery: "Show me all workspaces"

2. Selection: "List tests in the [Workspace name]"

3. Execution: "Run the [Test name] test with name 'API Performance Test - June 18'"

4. Monitoring: "What's the status of the running test?"

5. Analysis: "Analyze the results once the test completes"

1. Historical Data: "Show me the last 3 test results for [test name] from workspace [workspace name]"

2. Analysis: "Compare response times across these results"

3. Trend Identification: "Are there any performance degradation trends?"

"Show me all available workspaces"

"Find the workspace named 'MCP-demo'"

"Show the details for workspace [workspace name]"

"Show me all tests in the [workspace name] workspace"

With previous context, you can use shorter prompts:

"List all tests"

"Find all tests related to [test name] in workspace [workspace name]"

"What are the details of the [test name] test in workspace [workspace name]?"

"Run test [test name] from the workspace [workspace name]"

With test ID: "Run the test [test-id] with the name 'Performance Test - [Date]'"

"What's the status of the running test?"

"Stop the currently running test"

• Write as you would speak to a knowledgeable colleague

• Avoid overly technical jargon unless necessary

• Use conversational connectors ("Then", "Next", "Also")

Good examples:

• "Can you help me understand why our load test failed?"

• "I'm seeing some concerning trends in our API performance..."

• "Let's dig deeper into those error patterns."

• "What would you recommend for improving response times?"

Make your goals explicit rather than leaving them ambiguous:

• Avoid: "Do something with the data"

• Use: "I want to identify performance bottlenecks"

• Avoid: "Check performance"

• Use: "Analyze response time trends for the last 5 test runs"

• Avoid: "Why is it slow?"

• Use: "Why are the API response times in test result abc-123 three times slower slower than our baseline?"

• Avoid: "Analyze everything and tell me all the problems and solutions and comparisons and trends"

• Use: "Show me the tests that failed this week, and then help me investigate further"

• Avoid: "Fix the issue from yesterday"

• Use: "Help me troubleshoot the timeout errors in yesterday's load test for the payment API"

Guide the AI through your reasoning process step by step:

"I'm investigating a performance regression. Let's work through this step by step:
1. First, show me the recent test results for the user-service.
2. Then identify which metrics degraded compared to last week.
3. Next, analyze the error logs for those degraded tests.
4. Finally, correlate any errors with the performance drops."

Frame requests from specific perspectives:

"As a DevOps engineer preparing for a production release, analyze the staging environment performance tests and tell me if we're ready to deploy based on our SLA requirements."

Set clear boundaries and requirements:

"Generate a performance report that:
- Takes no more than 5 minutes to read
- Focuses only on critical issues (>10% degradation)
- Includes specific remediation steps
- Uses non-technical language for management review"