At the moment, I manage my knowledge base in Obsidian and use the Rules function to trigger updates via mcp when necessary. However, when the context becomes lengthy, these Rules often fail to run—for example, commands like “update the knowledge base when a task is completed” frequently don’t execute.
If we could hook into either the end of a thread or the completion of a single agent response, spin up an asynchronous chat in the background, and then run mcp, we could execute mcp in a more consistent and reliable way.