acp-install

Control the tool call, control the agent. See and stop what your coding agent actually does — every command, edit, and fetch — with one control layer that works the same across Claude Code, Cursor, and Codex. Free, on your machine, no account.

curl -sf https://agenticcontrolplane.com/install.sh | bash -s -- --local

One command. It detects whichever agents you run and puts the same guardrails in front of all of them:

• A safety floor nothing can cross — rm -rf /, mkfs, dd to a disk, a fork bomb, a force-push to main are blocked regardless of your policy (and regardless of how the command is spelled).
• Your rules, one file — ~/.acp/policy.json: allow / ask / deny per tool, applied identically to every agent.
• A log of what actually happened — tail -f ~/.acp/audit.jsonl and watch the calls your agent made, on-device. Nothing leaves your machine.

The same policy that stops Claude Code stops Codex. You configure control once, not once per vendor. Outgrow on-device? Re-run without --local to connect a workspace — team control, cost X-ray, and a shared console across everyone's agents. The local runtime is the free individual on-ramp; the cloud is the team upgrade.

Works on macOS + Linux. Requires Node 18+ and one of: Claude Code, Cursor, OpenAI Codex CLI, OpenClaw.

Want the long version first? Every file the installer writes, in plain language · getting started · per-client guides for Claude Code and Codex CLI

What the installer does

For whichever AI clients it detects:

1. Writes ~/.acp/govern.mjs — a shared hook dispatcher script that sends every tool call to the ACP governance API and enforces allow/deny decisions locally.
2. Registers PreToolUse + PostToolUse hooks in the client's config:
   • Claude Code: ~/.claude/settings.json
   • Cursor: ~/.cursor/hooks.json
   • Codex: ~/.codex/hooks.json
3. For Codex only — wires three layers:
   • Enables [features].codex_hooks = true in ~/.codex/config.toml
   • Adds [mcp_servers.acp] for non-Bash tool governance via MCP (with runtime credential substitution — no API key in your dotfiles)
   • Writes an ACP section in ~/.codex/AGENTS.md instructing Codex to call acp_check before non-Bash tool invocations
4. Opens a browser for OAuth to provision an ACP workspace and mint an API key
5. Saves the key to ~/.acp/credentials (mode 0600)
6. Installs the claude-acp cost-X-ray wrapper (~/.acp/bin) and adds one marked PATH line (# acp-installer) to your shell rc

In --local mode, steps 4–6 and every other piece of cloud wiring are skipped entirely: no OAuth, no MCP server, no cost wrapper, no shell-rc edits. Hooks + the on-device engine only. (Codex note: its hooks cover shell commands today; non-Bash Codex tools aren't hookable yet — in cloud mode the MCP connector covers them.)

The installer is idempotent: running it again upgrades existing entries in place without duplicating them or touching unrelated hooks/policies you've configured.

Trust signals

• Source: install.sh — one self-contained script, read it top-to-bottom
• SHA-256: https://agenticcontrolplane.com/install.sh.sha256 — auto-updates on every Agentic Control Plane release
• License: MIT
• Dry read: curl -sf https://agenticcontrolplane.com/install.sh | less
• Commit history: every change is here in this repo

The canonical install URL is agenticcontrolplane.com/install.sh (served from the marketing site). This repo is the auditable mirror.

What this script will NOT do

• Run any non-interactive commands without prompting if creds already exist (it asks "Reconfigure? (y/N)")
• Install to directories you don't own ($HOME/.acp/, $HOME/.codex/, $HOME/.claude/, $HOME/.cursor/ only)
• Phone home to any server other than api.agenticcontrolplane.com and (during auth) cloud.agenticcontrolplane.com — and in --local mode, nothing leaves your machine at all (decisions run on-device from ~/.acp/decide.mjs + ~/.acp/policy.json; no network calls)
• Modify anything outside the client config files documented above (cloud mode adds exactly one marked PATH line to your shell rc for claude-acp; local mode touches no rc files)
• Install binaries or compile anything — it's a pure shell + Node.js script

Uninstall

# Remove the ACP directory (credentials + govern.mjs)
rm -rf ~/.acp

# Remove the hooks from each detected client's config:
# - ~/.claude/settings.json        remove the "govern.mjs" entries under hooks.PreToolUse[] and hooks.PostToolUse[]
# - ~/.cursor/hooks.json           remove the "govern.mjs" entries under hooks.preToolUse[] and hooks.postToolUse[]
# - ~/.codex/hooks.json            remove the "govern.mjs" entries under hooks.PreToolUse[] and hooks.PostToolUse[]
# - ~/.codex/config.toml           remove the [mcp_servers.acp] block and [features].codex_hooks line
# - ~/.codex/AGENTS.md             remove the block between <!-- acp:begin --> and <!-- acp:end --> markers
# - ~/.zshrc / ~/.bashrc           remove the PATH line marked "# acp-installer" (cloud mode only)

A one-line uninstall.sh is planned. Until then, the blocks above are small enough to remove by hand.

Reporting issues

• Install broke something: open an issue here → github.com/agentic-control-plane/acp-install/issues
• Governance behavior questions: agenticcontrolplane.com/faq
• Integration details per client: agenticcontrolplane.com/integrations

License

MIT — see LICENSE.

Not a coding-agent CLI? Framework agents use acp-governance-sdks; Hermes Agent uses hermes-acp-plugin (pip install hermes-acp).