Skip to content
Agent kit

Run Crane from your terminal

npx crane is an MCP server and CLI that lets Claude Code or Codex drive your GTM engine in plain language — analyze a domain, generate plays, write sequences, prospect, and, with your approval, launch. The web app at cranereach.xyz stays your control room: everything the agent does is visible and editable there.

Install

You need Node.js and either Claude Code or Codex installed. From any project, register the MCP with your agent:

npx crane install

To register with only one agent, pass --claude-only or --codex-only.

Quickstart

Connect this machine to your Crane organization. A short code is printed in the terminal; you approve it as a signed-in human in the browser, so a stolen link alone can’t complete the flow.

npx crane login

That’s it. Now ask your agent in plain language — it uses the Crane tools to do the work:

  • Use Crane to analyze acme.com, show me 5 plays, and draft a trigger-led sequence for the best one.
  • Prospect 200 more leads for that list and enrich the ones missing an email.
  • Show me the campaign, then launch it once I approve.

The agent handles research and setup automatically. Anything that sends stops at your approval tier first — the AI does the work, you approve what goes out.

CLI commands

CommandWhat it does
crane installRegister the Crane MCP with Claude Code and/or Codex.
crane loginConnect this machine. A short code is printed here; you approve it in the browser.
crane whoamiShow the signed-in organization.
crane logoutForget the stored token on this machine.

Your credential is a personal access token bound to your org, stored on this machine. Revoke it any time under Settings → Agent Access.

Tool reference

Once installed, your agent can call these Crane tools. Research, staging, and reads are reversible and run automatically; sending is approval-gated (see below).

Research & staging

Reversible by default — your agent runs these on its own to do the legwork. Nothing sends.

  • crane_analyze_domainRead a company site and build its GTM strategy + ideal-customer profile, saved as your knowledge base.
  • crane_generate_playsGenerate five signal-based outbound plays for a domain — each with the buying signal, the angle, and reach.
  • crane_list_strategiesList the available messaging strategies (trigger, pain, social proof, insight, direct).
  • crane_preview_sequenceDraft email-sequence copy for an angle without saving. Compare one variant per strategy side by side.
  • crane_run_playStage a full campaign from a play — ICP, campaign, sequence, and list — and start finding leads. Does not send.

Prospect & enrich

Fill and verify your lists so what reaches your table is worth working.

  • crane_get_more_leadsFind more matching companies and contacts for a list.
  • crane_enrich_leadEnrich a lead with verified contact details.

Read & observe

Let your agent report on what is running — read-only, always safe.

  • crane_homeDashboard metrics and a recent-activity summary for your org.
  • crane_list_campaignsList your campaigns.
  • crane_get_campaignGet one campaign by id.
  • crane_list_sequencesList the sequences for a campaign.
  • crane_get_leadsList the leads in a list.
  • crane_permissionsShow your org’s agent approval tiers.

Send

Approval-gated. When your tier is set to “ask”, the agent pauses for a human before anything goes out.

  • crane_launch_campaignLaunch a campaign — this begins sending. Honors your approval tiers; requires confirmation when gated.

Approvals & control

Crane is built so the AI does the work and you approve what matters. Research, list-building, and drafting run on their own because they’re reversible. Irreversible actions — anything that sends — pass through approval tiers you configure per organization.

When a tier is set to ask, the agent pauses and waits for a signed-in human to approve before it proceeds. Set it to allow for a hands-off flow, or keep it on ask for a review gate on every send. Either way, every list, sequence, and campaign the agent touches is visible and editable in the web app — nothing happens off-screen.