What
MCP servers let Claude Code use outside tools. A stdio server is one Claude Code starts as a local program and talks to through its input and output. When connecting, Claude Code sends a server/discover probe to learn whether the server speaks the newer protocol. A server that does not answer in time is treated as a legacy (old-protocol) server and remembered that way.
- A stdio server whose probe timed out is now tracked (
_anthropicProbeTimedOut,answeredProbeLate). If it answers late, it counts as slow to start rather than old, and is not recorded as a legacy server. - If it answers after it has already connected as legacy, a callback (
onAnsweredProbeLate) forgets that legacy record. - The condition that sends a connection down the legacy path was widened from one flag to
(Ie || er). - Servers configured with a
commandnow get an identity: an id hashed from the command, its arguments, its environment variables in sorted order and a project folder. Before, only http servers got one.
One finding links this to the built-in default of tengu_mcp_protocol_negotiation_stdio moving from off to on; the other could not confirm that link. Readings of that gate came from the flag server, which returned on for this site's account and for the anonymous baseline, switched on since the previous reading; no reading has been taken under this release yet.
Why
A stdio MCP server that is simply slow to start no longer gets stuck on the old protocol path.
tengu_mcp_protocol_negotiation_stdio On for this account, and not off by defaultThe flag server returned on for the one account this site reads, and nothing in this release compiles it off by default. The compiled default is shown below, and says which it is when we cannot read one: a fifth of gates compile in a string or a number rather than on or off, and most published releases have no gate table behind them at all. No client can see what the server returns for your account.
This account: on · anonymous baseline: on · compiled default in v2.1.292: 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 v2.1.292. It isn't a statement about your account. What a flag value here can and cannot tell you
A reading is one sample. Claude Code evaluates its flags remotely, so no client sees the targeting rule behind a value and this says nothing about your account.
A reading is one sample. Claude Code evaluates its flags remotely, so no client sees the targeting rule behind a value and this says nothing about your account.