Sign in to an MCP server mid-session and the next turn still comes back with none of that server's tools.
openclaw mcp login <name>
# sign-in succeeds, and the next turn runs without the server's tools
The turn after that works. It looks like the login did not take, so people sign in twice. This is openclaw/openclaw#172614, fixed in PR #172654.
Why it happens
When a bundled server fails to start - it needs OAuth and has no credentials yet, or its transport refuses once - the catalog owner stores a failure catalog and schedules a retry. While the retry is pending it serves the stored catalog and recovers in the background. For a server that worked before, that is the right trade. For a server that has never listed in this runtime, the stored catalog is the failure, with none of its tools, so the first turn after the retry comes due runs against a catalog that is already wrong.
Signing in does not help on its own. The login does not change the server's config fingerprint and does not touch the retry schedule, so nothing invalidates the cached failure. The tools are sitting right there. The session just has not looked again.
The fix
The first turn after a retry is due now waits for the catalog loader until that server has listed successfully, instead of returning the cached failure and recovering in the background. If startup still fails, the catalog stays empty for that server and reports its unavailable reason, the next retry time, and reachability guidance. Healthy servers keep their existing background recovery.
On affected versions, expect one turn of missing tools after openclaw mcp login <name>. If you cannot wait, restarting the session picks the tools up immediately. Upgrade for the fix.