Claude Code Updates: Faster Streaming, Reliable MCP, Self-Healing Sessions

Here’s a list of updates on everything that was recently improved in Claude Code:

1. First, for the new fullscreen renderer (which is meant to eliminate bugs like screen flickering), they made many fixes for different environments and terminals.

You can enable it with the command:

/tui feedback

They’re currently working on making it the default mode in Claude Code in the near future.

2. Claude’s responsiveness during work has been significantly improved. Now the reasoning process and tool calls are streamed in real time, and they also fixed a number of bugs that could make it seem like Claude had frozen, when in fact it was just taking a long time to process the request.

3. There are fewer mysterious error messages. They found and fixed several root causes of errors like "tool result doesn’t match tool use", and also made other error messages clearer so it’s easier to understand what exactly is happening.

4. Compaction now shows execution progress, and you should no longer run into bugs like "prompt too long" that previously could completely block compaction.
Anthropic is also gradually rolling out additional changes to speed up the compaction process itself in Claude.

5. MCP has become much more reliable. They released a set of fixes for connection issues, the OAuth flow, and proxy-level rate limiting.

6. Self-healing sessions. Previously, overly large or unreadable images and media files could completely break a session until a restart. Now Claude Code automatically detects such situations and restores operation on its own.

7. Now when calling /feedback, you can simply send sessions from the last day or week, instead of trying to remember which exact session the bug happened in.