### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) and this hasn't been reported yet - [x] This is a single bug report (please file separate reports for different bugs) - [x] I am using the latest version of Claude Code ### What's Wrong? Starting the morning of 2026-06-22, every Claude Desktop SSH session to a remote instance fails immediately. The UI shows: > Try sending your message aga

[BUG] Desktop SSH: ccd-cli 2.1.181 "is not runnable" — versioned dir never created on a shared remote instance with multiple Claude accounts

Open 💬 0 comments Opened Jun 22, 2026 by mshih2

Preflight Checklist

[x] I have searched existing issues and this hasn't been reported yet
[x] This is a single bug report (please file separate reports for different bugs)
[x] I am using the latest version of Claude Code

What's Wrong?

Starting the morning of 2026-06-22, every Claude Desktop SSH session to a remote
instance fails immediately. The UI shows:

Try sending your message again. If it keeps happening, share feedback so we can investigate.

~/Library/Logs/Claude/main.log shows the underlying error repeating on every
preconnect / getSSHGitInfo / session warm attempt:

Failed to install Claude CLI: installed cli at /home/nvidia/.claude/remote/ccd-cli/2.1.181 is not runnable
    at J5e (/Applications/Claude.app/Contents/Resources/app.asar/.vite/build/index.js:3285:3893)
    at process.processTicksAndRejections (node:internal/process/task_queues:104:5)
    at async TQn (...)
    at async cu.doEnsureReady (...)

The remote ccd-cli/2.1.181/ directory is never successfully created — the
install step builds the parent path but leaves no runnable binary, so the
post-install runnability check fails.

This worked fine until this morning; it appears to coincide with an auto-update
to CLI/ccd-cli 2.1.181.

Environment

Claude Desktop (macOS): 1.14271.0
ccd-cli version being installed: 2.1.181
Local machine: MacBook Air, macOS <FILL IN: e.g. 15.x>, <FILL IN: Apple Silicon / Intel>
Remote: NVIDIA Brev GPU instance (hostname brev-952f0g62)
Remote OS: Ubuntu, glibc 2.39 (Ubuntu GLIBC 2.39-0ubuntu8.7), x86_64
Remote user: nvidia
Important: This single remote instance is **shared by 3 people, each signed

in to a different Claude account.**

Steps to Reproduce

Have a remote host that 3 users share, each authenticating to a different

Claude account, all landing in the same remote ~/.claude (Desktop's SSH
integration offers no way to set a per-user remote config dir).

From Claude Desktop (macOS) → Code tab, connect to that host over SSH and

select a project folder.

Send any message (or just let the session warm).
Observe the generic "Try sending your message again" UI error; main.log

shows installed cli at /home/nvidia/.claude/remote/ccd-cli/2.1.181 is not runnable.

What I Ruled Out (with evidence)

The remote environment is healthy — this is not a missing-loader, disk,
network, or architecture problem:

| Check | Result |
|---|---|
| uname -m | x86_64 |
| ldd --version | Ubuntu GLIBC 2.39-0ubuntu8.7 |
| glibc loader /lib64/ld-linux-x86-64.so.2 | present (symlink to ../lib/x86_64-linux-gnu/...) |
| musl loader /lib/ld-musl-x86_64.so.1 | present (symlink to x86_64-linux-musl/libc.so) |
| df -h ~ | 75G free (85% used) — not full |
| curl -sSI https://downloads.claude.ai/claude-code-releases/latest | HTTP/2 200 — egress OK |
| ls -la ~/.claude/remote/ccd-cli/2.1.181/ | No such file or directory — versioned dir never created |
| ~/.claude/remote/run/*/remote-server.log (newest) | dated 2026-06-21; no 2026-06-22 activity — remote server never started today |

Local native binary is also healthy and runs fine:

$ ls -la "~/Library/Application Support/Claude/claude-code/2.1.181/claude.app/Contents/MacOS/claude"
-rwxr-xr-x@ 1 user staff 215193056 Jun 18 01:24 .../claude
$ "~/Library/Application Support/Claude/claude-code/2.1.181/claude.app/Contents/MacOS/claude" --version
2.1.181 (Claude Code)

Misleading error message in main.log

main.log also emits this, which sent debugging in the wrong direction:

Claude Code native binary at /Users/<user>/Library/Application Support/Claude/claude-code/2.1.181/claude.app/Contents/MacOS/claude exists but failed to launch. This usually means the binary does not match this system's libc — e.g. spawning a musl-linked binary on a glibc Linux host fails because the musl dynamic loader (/lib/ld-musl-*) is missing. Specify a matching binary with options.pathToClaudeCodeExecutable.

This message is misleading here: the referenced local binary runs fine
(--version → 2.1.181), and on the remote the musl loader is present
anyway. The "musl on glibc / missing loader" explanation does not apply to
either machine, yet it is surfaced as the likely cause. Suggest making this
error conditional on an actual loader/libc check rather than printing it as a
generic catch-all on any launch failure.

Suspected Root Cause

I suspect this is triggered by multiple Claude accounts sharing one remote
~/.claude, not by a plain 2.1.181 packaging defect:

Desktop's SSH integration hardwires the remote config to ~/.claude and

exposes no field to set a per-account remote config dir.

With 3 users on different accounts all writing into the same remote

~/.claude/remote, credential / install state can collide, which plausibly
leaves the ccd-cli/<version>/ install in a half-finished, non-runnable state.

This would explain why wiping ~/.claude/remote and reconnecting does not

help (the next connecting account re-collides), and why the versioned dir is
built but left empty/absent.

Impact / Why It Matters

Desktop SSH is effectively unusable for any multi-user shared remote where

users have distinct Claude accounts.

The only error shown in the UI is a generic "try again", with the real cause

buried in main.log — please surface the actual install/launch failure in the
UI (this is a recurring theme across related issues).

Workarounds Tried (none resolved it)

Full quit + restart of Claude Desktop — no effect.
Deleting ~/.claude/remote / the ccd-cli/2.1.181 dir and reconnecting — dir

is rebuilt but binary still never lands; same error.

Requested Fixes

Allow specifying a per-connection remote config dir (e.g. honor

CLAUDE_CONFIG_DIR) for Desktop SSH, so users sharing a host can isolate
accounts. (Related: existing request to make the .claude config dir
relocatable.)

Surface the real ccd-cli install/launch failure in the UI instead of the

generic "try again".

Gate the "musl-linked binary on a glibc host / missing loader" message behind

an actual loader check; it is currently emitted misleadingly.

What Should Happen?

Claude code sessions through Claude desktop should be runnable.

Error Messages/Logs

Something went wrong
Try sending your message again. If it keeps happening, share feedback so we can investigate.
Failed to install Claude CLI: installed cli at /home/nvidia/.claude/remote/ccd-cli/2.1.181 is not runnable

Steps to Reproduce

Open Claude Desktop and go to Claude C

ode sesssions

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.181

Platform

Other

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_

View original on GitHub ↗