How to recover a Claude Code session after VS Code crashes or the terminal closes

You were in the middle of something delicate. Claude had the shape of the refactor in its head, the last few edits made sense, and you were finally past the awkward setup phase where you have to keep re-explaining the repo. Then VS Code freezes. Or the integrated terminal disappears. Or you close the window without thinking and realize a second later what you just lost.

When you restart, the code is still there, but the session is gone. The half-formed reasoning, the next step Claude had lined up, the little thread connecting what you just changed to what you were about to change next, all of it is missing. Now you are not continuing work. You are reconstructing it.

There is a name for this feeling: session discontinuity. The code survives the crash, but the working context does not, and those are not the same thing. A repo can be intact while your momentum is completely broken.

What makes this frustrating is that the real loss is usually not the last few lines of code. Git, autosave, and your editor cover more of that than people think. What disappears is the short-lived context that lived inside the session itself. Why Claude was looking at those files. Which path you ruled out five minutes ago. Whether the next move was to add the test first or finish the migration first.

That context is fragile because Claude Code sessions are not durable by default. They are great while they are alive, but a crash or restart cuts the thread unless something outside the session captured it. So when people say “I lost my work,” what they often mean is “I lost the state that made the next 20 minutes obvious.”

The natural fix is not trying to remember harder after the restart. It is making sure the session leaves behind enough state that the next one can pick it up. That is what KeepGoing’s MCP server is for. When you run npx @keepgoingdev/mcp-server, Claude can read your saved checkpoint history and session context at the start of a new session. Instead of reopening VS Code and staring at the prompt trying to remember what just happened, you get a re-entry briefing with what you were doing, what changed, and what to do next.

That matters most after an interruption you did not plan for. A crash turns from “great, now I have to rebuild the whole mental stack” into “okay, here is where I was.” You are not asking Claude to guess from the diff. You are giving it the thread back.

The next time a session gets cut off, do one small thing before you dive back into the code: start the new session with npx @keepgoingdev/mcp-server configured and pull the re-entry briefing first. If a terminal recording shows the interruption and the resume side by side, the difference is obvious immediately. One restart begins with archaeology. The other begins with momentum.