Miniviz MCP Tips
Miniviz MCP works best when your question identifies a project, a time range, and the metric or condition you care about. Start broad, then narrow the question with the result.
Start With Context
If the project is not clear, ask:
Which Miniviz projects can I use in this chat?
Then identify the project by name in later questions.
Check the Latest State
Use this for a quick current-status check:
For <project>, show the latest temperature and humidity readings.
Name the fields you need. When field names are unknown, ask:
For <project>, what event fields and label values are available?
Ask for a Bounded Trend
Make the period and aggregation explicit:
For <project>, summarize temperature over the last 24 hours in one-hour averages.
For longer periods, ask for coarser buckets:
For <project>, show the monthly temperature trend for the last year.
Count Conditions Before Inspecting Events
For threshold questions, ask for the count first:
For <project>, how many events had temperature at or above 30 during the last 30 days? Show the daily top five.
Then inspect a narrow period if you need individual records:
Show the matching temperature and status events on the busiest day.
This sequence is clearer than asking for a large raw-event dump.
Compare Labels Deliberately
When a project has multiple devices or locations, name the label values you want to compare:
Compare the last 24 hours of temperature for label values <label-a> and <label-b> in <project>.
Ask for Explanation, Not Just Values
Miniviz MCP can also help you interpret configured monitoring:
For <project>, explain the configured charts and monitoring rules in plain language.
It returns safe summaries and does not reveal notification destinations or credentials.
Know When to Use the App Instead
Use the Miniviz Database page when you need a dense table view, manual CSV work, or broad row-by-row inspection. MCP is intentionally bounded for conversational analysis and does not provide unlimited raw-event export.
Miniviz MCP also does not calculate exact continuous durations, detect every missing-data interval, or write data and settings. For these cases, narrow the question, inspect the data in the app, or use your own downstream analysis workflow.