We hit a practical footgun while using the CodeGraph MCP Code Context query.
A vague natural-language query like:
Find async execution, message queue, Spring Cloud Stream event publishers and consumers
can drift toward unrelated symbols such as streamChat / stream, because the query does not give CodeGraph enough concrete code anchors.
Could the installer add a very short memory note for the AI about how to write Code Context queries?
Suggested memory text:
When using CodeGraph Code Context, include concrete symbols, class names, method names, annotations, or framework keywords in the query. Avoid vague natural-language-only queries. Example: EventPublisher Consumer DomainEvent StreamBridge @Async AiTaskExecutorService AsyncConfig.
This would help agents use the MCP more reliably right after install.
We hit a practical footgun while using the CodeGraph MCP
Code Contextquery.A vague natural-language query like:
can drift toward unrelated symbols such as
streamChat/stream, because the query does not give CodeGraph enough concrete code anchors.Could the installer add a very short memory note for the AI about how to write Code Context queries?
Suggested memory text:
This would help agents use the MCP more reliably right after install.