Shows how to hook Cline up to an MCP research tool so it gathers current source-backed context before it touches code, which should save a few “looks fine, breaks later” moments.
“Research before touching code” is a nice idea, but I can already picture the agent doing the digital version of opening 30 tabs just to change a button color.
How are you deciding when it’s allowed to do the deep crawl vs when it should stick to repo context + whatever docs you already have locally? I’m mainly wondering what your trigger is (new dependency, suspected breaking API change, flaky/weird bug, etc), because otherwise this feels like it could turn into a 5‑minute detour on every tiny PR.
