Skip to main content
The claude-code runner invokes the claude CLI binary as a subprocess rather than using the SDK in-process. This produces the same model output as claude-sdk but runs in a separate process, which can be useful for isolation, reproducibility, or when you want full control over the claude binary version independently of the RemoteAgent package.

​
Requirements

  • The claude CLI installed globally: 
    npm install -g @anthropic-ai/claude-code
  • An Anthropic API key. The claude binary reads it from the same local config RemoteAgent uses, or from the ANTHROPIC_API_KEY environment variable.
  • Node.js 20 or later.
Verify the binary is available: 
claude --version

​
Setup

remoteagent init --runner claude-code
Or choose claude-code when prompted by the interactive wizard.

​
Difference from claude-sdk

Both runners use the same underlying Claude model and produce equivalent output. The key differences are:
claude-sdkclaude-code
ExecutionIn-process (Node.js)Subprocess (claude binary)
Startup overheadNone~200–500ms per invocation
Process isolationShared with agentSeparate OS process
Binary version controlTied to SDK version in packageIndependent — update with npm update -g @anthropic-ai/claude-code
Error outputStructured SDK errorsRaw CLI stderr
For most users, claude-sdk is the better choice. Use claude-code if you specifically need process isolation or want to pin the claude binary to a particular version.

​
How it works

The runner executes: 
claude -p "<prompt>" --dangerously-skip-permissions
The --dangerously-skip-permissions flag is required because the agent runs non-interactively — there is no user present to approve individual tool calls. RemoteAgent runs this on your own machine, so the permissions boundary is the same as running claude manually. Output is captured from stdout and streamed back to the Redis output channel chunk by chunk.

​
Updating the binary

npm update -g @anthropic-ai/claude-code
claude --version
Restart the agent after updating: 
remoteagent start