@anthropic-ai/claude-code SDK fails with "spawn node ENOENT" in Docker containers despite working CLI

using spawn, I was able to get it to work only with stdio: ['ignore', 'pipe', 'pipe']

ignoring the stdin was the trick :)

I'm hitting this error outside docker container, anyone else able to fix this?

If this sdk doesn't work inside of a container, i struggle to understand what it's even for?
Does anyone use it in production?

Had to make sure to pass my path to env, since it failed to resolve node.

    const agentQuery = query({
      prompt: task,
      options: {
        model: 'claude-sonnet-4-5-20250929',
        maxTurns: 20,
        cwd: process.cwd(),
        env: {
          PATH: process.env.PATH,
          ANTHROPIC_API_KEY,
        },
      },
    });

I encounter abnormal exit issues in the following scenarios, for reference:

1. The specified path for cwd does not exist

Solution: Fix the relevant program logic

1. Under the Docker Alpine image, Bash is specified in allowedTools

query({
   allowedTools: [
      Bash,
      BashOutput,
      KillBash
   ],
})

Solution: Add the RUN apk add bash command to the Dockerfile

1. The bypassPermissions setting is configured in Docker

The bypassPermissions mode requires sudo privileges.

Solution: Change permissionMode: bypassPermissions to permissionMode: default to resolve the issue

Debugging Tips: By setting the DEBUG=* environment variable, you can view the runtime logs of Claude Code, which helps you better understand the causes of abnormal exits.

from https://lzw.me/a/claude-agent-sdk-exited-with-code-1.html

I'm encountering this same issue in a Vercel Sandbox environment. I've verified that:

• The Claude Code CLI is properly installed in the container
• The PATH environment variable is correctly configured and passed to the query() call
• The CLI works when invoked directly from the command line

However, the SDK continues to fail with the spawn node ENOENT error. I also tried hardcoding pathToClaudeCodeExecutable to the Claude Code CLI path in my container to no avail.

@ashwin-ant, could you provide an update on the status of this issue? Since this was reported in July, I wanted to check if there's a fix planned or if there's a recommended workaround in the meantime. This limitation is currently blocking our ability to integrate the SDK into containerized/production environments, so any guidance would be greatly appreciated.

If there are any additional diagnostic steps or configurations I should try, I'm happy to help test potential solutions.

@josh-d2 can you share more about your setup? That is, what sort of options are you passing into query? Are you running as root or using any specific permission mode?

@ashwin-ant, I discovered this issue while following this Vercel tutorial for running Claude's tooling in a Vercel Sandbox.

Key difference that triggers the bug:

1. The tutorial directly invokes the Claude Code CLI (@anthropic-ai/claude-code) and works fine
2. My application uses the Claude Agent SDK (@anthropic-ai/claude-agent-sdk) as a dependency, which then invokes Claude Code internally
3. My app works locally but fails with the spawn node ENOENT error in the sandbox environment

This suggests the issue is specifically with how the Agent SDK spawns the Claude Code process in containerized environments, rather than with the Claude Code CLI itself.

Interesting note: The Vercel tutorial is titled "Using Vercel Sandbox to run Claude's Agent SDK" but actually demonstrates using the Claude Code CLI directly, not the Agent SDK. I wonder if the tutorial author originally attempted to use the Agent SDK, encountered this same spawn issue, and pivoted to using the CLI directly—but forgot to update the tutorial title.

This would explain why:

1. The tutorial works (it's using the CLI directly)
2. My implementation fails (it's using the Agent SDK, which has the spawn bug)
3. The title is misleading (it claims to show Agent SDK usage)

This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.

This issue has been automatically closed due to 60 days of inactivity. If you're still experiencing this issue, please open a new issue with updated information.

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.