MCP servers are outside programs that give Claude Code extra tools. Claude Code now has a separate error for two cases:
- a reply from an MCP server is larger than the size limit
- a reply could not be read and the connection closed
The error tells Claude: "The tool may have run: check whether it did before running it again." When the connection closes this way, Claude Code no longer automatically reconnects, signs in again and retries.
A connection that fails because a request timed out is still marked as a connection timeout. That marking sits behind a remote switch that falls back to on when the server sends nothing.
Some tools should not run twice, such as one that sends a message or makes a payment. Retrying blindly after an unreadable reply could repeat an action that already happened, so Claude is now told to check first.
tengu_mcp_connect_timeout_retry Not enough to sayNothing here resolved what this flag was doing on this version, so nothing here should be read as on or off.
This account: no value returned · anonymous baseline: no value returned · compiled default in v2.1.288: on
These values were read against a different version of Claude Code, so treat them as the nearest reading available instead of one taken on this release.
Read once, for one account on one subscription tier, against v2.1.288. It isn't a statement about your account. What a flag value here can and cannot tell you
It is not clear exactly which MCP calls can raise the new error.