Querying Memory with AI

One of the most powerful features of the MCP-U client is the ability to let AI models (like Claude or Gemini) query historical sensor data themselves. They do this using the sql_readonly_query tool.

Here are some tips and example prompts for getting the most out of this feature.

1. System Prompts & Context

The AI doesn’t automatically know your database structure. However, the client provides a resource at mcu://cache/schema.

Effective Prompt:

“Read the schema at mcu://cache/schema to understand the database structure, then write an SQL query to find the average voltage over the last 10 minutes.”

2. Example Use Cases

Detecting Anomalies

“Using the sql_readonly_query tool, look at the numeric_observations view for the resource voltage_feedback. Are there any values above 3.5V in the last hour? If so, tell me exactly when they occurred.”

Summarizing Tool Executions

“Query the mcp_u_tool_calls table and give me a summary of how many times the set_led tool failed versus succeeded today.”

Poller Health Check

“Check the buffer_drain_health view. Are there any error counts for the buffer drains? Are we pulling data consistently?“

3. SQL Safety

The sql_readonly_query tool is equipped with a safeguard. It only permits SELECT and WITH statements. If an AI tries to run an INSERT or DROP TABLE command, the client will block the query and return an error.

The tool also enforces a maximum row limit (default: 500) to prevent the AI context window from being overwhelmed by massive datasets.

If the AI needs to process more data, prompt it to use SQL aggregations instead:

“Do not return all the raw data. Instead, write an SQL query that groups the data by hour and returns the AVG(value_num).”