claudecmd

pip install pyclacmd → the claudecmd command. Script Claude Code from the shell while keeping calls on your interactive (subscription) path, not separately-priced -p/headless usage.

A small, reliable command that drives your local, already-installed Claude Code interactive session programmatically — from scripts, shell pipelines, and editor integrations — and returns the assistant's reply as plain text or a stable JSON envelope.

It drives the interactive session (not claude -p) on purpose: Claude Code prices -p/headless usage separately from interactive sessions, so claudecmd keeps automated calls on your interactive (subscription) path. There is no network server and no API key handling of its own — it shells out to the claude binary you already use, under a pseudo-terminal.

The PyPI package is pyclacmd; the installed command is claudecmd.

claudecmd "say hello"
echo "summarize this" | claudecmd
git diff | claudecmd "Review this diff for risky changes"
claudecmd --json "explain this repo"
claudecmd --session 8f3c… "continue from previous context"

Demo

---

Requirements

• macOS (primary target; also works on Linux).

• Python 3.8+.

• Claude Code installed and authenticated. claudecmd invokes the claude binary found on your PATH. Validate your setup:

  command -v claude
  claude --version
  claude "say hello"      # interactive session works for your login

  Authentication (claude /login or a subscription token) is a prerequisite — claudecmd does not manage it. Point claudecmd at a specific binary with the CLAUDECMD_CLAUDE_BIN environment variable.

---

Install

pip install pyclacmd

This provides the claudecmd console script. pexpect and pyte (used to drive and render the interactive TUI) are installed automatically.

macOS note: if the system pip is too old, use a virtualenv:

python3 -m venv .venv && source .venv/bin/activate
pip install pyclacmd

---

Usage

claudecmd [options] [prompt]

The prompt comes from (in priority order): the positional argument, then piped STDIN. If both are present they are combined:

<positional prompt>

--- STDIN ---
<stdin content>

Flags

Flag	Description
--json	Emit a stable JSON envelope (see below).
--raw	Print the full rendered TUI screen (debugging aid).
--cwd <path>	Working directory Claude runs in.
--session <id-or-name>	Resume a session by UUID or local name.
--timeout <seconds>	Abort (and clean up) after N seconds.
--model <model>	Model alias (opus, sonnet, haiku) or full id.
--tools <tools>	Built-in tools to allow (e.g. "Bash,Read"); "" disables all.
--permission-mode <mode>	Claude permission mode (e.g. plan, acceptEdits).
--system-prompt <text>	Replace the system prompt.
--append-system-prompt <text>	Append to the system prompt.
--allowed-tools <patterns>	Permission allow patterns, e.g. "Bash(git:*),Read".
--disallowed-tools <patterns>	Permission deny patterns.
--add-dir <path>	Extra allowed directory (repeatable).
--debug	Emit redacted diagnostics to stderr; keep oversize-stdin temp files.
--dry-run	Print the command plan and exit without calling Claude.
--version / --help	Standard.

Examples

# Basic
claudecmd "explain this repo"

# Pipe input as JSON
cat task.md | claudecmd --json

# Review a git diff
git diff | claudecmd "Review for risky changes and return concise findings"

# Unattended: disable tools so no permission prompt can stall the run
claudecmd --tools "" "summarize the open questions in this file"

# See exactly what would run, without running it
claudecmd --dry-run "say hello"

---

Output contract

Default (human) mode

Prints only the assistant's final reply to stdout, extracted from the rendered session. No banners, no metadata.

--json mode

Emits one JSON object:

{
  "ok": true,
  "result": "assistant response text",
  "session_id": null,
  "duration_ms": 6824,
  "cost_usd": null,
  "raw": null,
  "mode": "interactive"
}

session_id, cost_usd, and raw are null — the interactive TUI does not expose them. On failure:

{ "ok": false, "error": "…", "kind": "claude_timeout", "exit_code": 124, "duration_ms": 1234 }

--raw mode

Prints the full rendered TUI screen unchanged — useful for debugging extraction.

--dry-run mode

Prints the exact command plan (run through the secret redactor) and exits:

{ "ok": true, "dry_run": true, "command": ["claude", "say hello", "--model", "haiku"] }

---

How it works

claudecmd spawns claude "<prompt>" (interactive, no -p) under a pseudo-terminal, renders the TUI with a real terminal emulator (pyte) so layout and whitespace survive, auto-answers the one-time workspace-trust dialog for --cwd, waits for the turn to settle, and extracts the assistant's reply from the rendered screen.

Caveats — it scrapes a human-facing TUI, so it is inherently less robust than a headless API:

• No session_id or cost_usd is available (the TUI does not expose them).
• Completion is detected heuristically (the reply settles and the input box returns). Give long replies a larger --timeout.
• A tool-permission prompt will stall an unattended run — pass --tools "" to disable tools, or an appropriate --permission-mode.
• Extraction is tuned to Claude Code's current TUI (v2.1.x) and may need updating if the interface changes.

---

Sessions

• A --session value that is a UUID is resumed directly (--resume).
• A friendly name is looked up in ~/.claudecmd/sessions.json (override the directory with CLAUDECMD_HOME) and resumed if present. Note: because the interactive TUI does not surface the session id, claudecmd cannot record a new name→id mapping on this path — pre-seed names or resume by UUID.

---

Large input

STDIN larger than the guard (default 9 MB) is not silently truncated. It is written to a restrictive (0600) temp file under the OS temp directory, and the prompt references that file path so Claude can read it. The temp file is deleted after the run unless --debug is set.

---

Error kinds

Every failure carries a stable kind (the kind field in --json mode; in the stderr message otherwise) and a non-zero exit code:

Kind	Exit	Meaning
claude_not_found	127	The claude binary was not found / not executable.
claude_auth_required	2	Claude reported an authentication problem.
claude_timeout	124	The run exceeded --timeout.
cwd_not_found	66	--cwd is not an existing directory.
session_store_error	74	The session mapping file could not be read/written.
pty_unavailable	69	pexpect/pyte could not be imported.
no_prompt	64	No prompt via argument or stdin.
unknown	1	Anything else (including no extractable reply).

---

Security notes

• Prompts and flags are passed as a subprocess argument array — never interpolated into a shell string, so there is no shell-injection surface.
• Prompts are not written to disk except for the oversize-stdin case above.
• Temp files and the session store use restrictive permissions.
• --debug output and the --dry-run command plan are run through a best-effort secret redactor (API keys, bearer tokens, Authorization: headers, .env-style assignments).
• No permission-bypass flags are enabled by default. --permission-mode is a passthrough: explicitly passing bypassPermissions opts into Claude Code's auto-approve behavior — a deliberate choice, never a default.

---

Development

pip install -e ".[dev]"
pytest

The test suite mocks the claude binary / interactive runner, so CI never requires a real Claude login. GitHub Actions runs pytest on Python 3.12 / Ubuntu; the package itself supports Python 3.8+.

License

MIT — see LICENSE.