So I've been playing around a TON with MCP, @roocode and @openrouter 's Quasar Alpha stealth model for the last 36 hours or so. I think I've unlocked a new level.
Here is my setup;
- RooCode primarily in "Boomerang Mode" using Gemini
- Code mode using Quasar
- Architect mode using Gemini
- MCP servers for; project-memory, global-memory, web-search, sequential-thinking, sqlite-db, and filesystem
Custom instructions to keep a running set of implementation details in project-memory. Remember any of user's stated preferences using global-memory, and to use web-search to check up-to-date documentation at the first sign of a framework not behaving as expected (this helps a TON with svelte 4 runes). And to verify that the original user request is actually solved before marking a task as complete.
This has been my setup for the last 10 hours or so - most of today. This is easily the smoothest experience I have ever had using LLMs to code so far.
I think my next step is to make or find a https://t.co/0IdvaCUs7V MCP server and just tell Roo to pick the most important linear task based on highest priority and nearest due date.
https://t.co/aSjojTzq5u
@kathylienfx $EUR is, after many missed forecasts, a hard sell in the medium-to-long term however inflation has edged up last month and today’s PMIs seem to indicate that the European economy is not doing THAT bad. Anyway, we could see the current small rally evaporate in a matter of days.