AI
Claude Code 2.1.311 keeps going past 200K tokens instead of stopping
The Oct 26 release notes on GitHub also let helper agents use late-connecting tools.
Disclosure: Anthropic makes Claude, the model that helps make this paper. Version 2.1.311 of Claude Code, released Oct 26, changes what happens when a session runs past 200K tokens (chunks of text) on standard context. It now compacts (summarizes the older conversation) and continues, instead of stopping with "Usage credits required for 1M context."
Subagents (helper agents the main one hands jobs to) can now use MCP tools (plug-in tools an agent can call) from servers that finish connecting after the subagent starts. The caveat: our link is the releases list page, not a version-specific page.
On your desk: long vault sessions with your Claude agents should stop hitting that hard wall, and MCP-heavy subagent runs should be steadier.
Read the original at GitHub: anthropics/claude-code releases (v2.1.311, 26 Oct) →