Why Scout requires Claude Code

Scout isn't a standalone desktop app — it's an agent harness that runs inside Anthropic's Claude Code CLI. To run Scout, you need Claude Code installed and an active Anthropic plan (Pro, Max, or API credits). You pay Anthropic directly; we don't proxy or resell tokens.

What this actually means

Most "AI sales tools" are SaaS products that wrap a chat-completion API. You sign up, upload your CRM, and they run prompts against OpenAI or Anthropic on a server you don't control. Your prompts, leads, and replies pass through their infrastructure.

Scout is different. Scout ships as a set of agent skills that run inside Claude Code on your laptop. When Scout finds leads, scores them, drafts a message, or sends a connection request, that work is happening on your machine — Claude Code is the runtime, Scout is the playbook.

Concretely:

• You install Claude Code with your Anthropic credentials.
• You install Scout, which adds Scout's agent skills, MCP servers, and tools to Claude Code.
• Scout runs by invoking Claude Code — every plan, every search, every message draft is a Claude Code session you can watch in real time.

Why Claude Code, specifically

We considered building Scout as a standalone app calling the Anthropic API, the OpenAI API, or local models. We didn't because:

• Real agent loops, not request/response. Sales prospecting isn't one prompt — it's research, planning, scraping, scoring, drafting, retrying. Claude Code's agent harness handles multi-turn tool use, sub-agent orchestration, and long-running tasks reliably. Building that ourselves would be a year of infrastructure for a product the size of Scout.
• The model is currently the best at agentic work. Anthropic's Claude Sonnet/Opus 4.x consistently lead the agentic benchmarks (SWE-Bench, OS-World, Browser-Use). For Scout's specific failure modes — sub-agents that lose context, tool use that hallucinates parameters, plans that derail mid-task — Claude is just measurably better right now. If that changes, we'll revisit.
• Skills as a real format. Claude Code's skill system lets us ship Scout as a set of inspectable, modifiable Markdown files instead of a black-box binary. You can read every prompt Scout uses, fork them, override them, or wire in your own tools.
• It's already running locally. Claude Code is a CLI tool that runs on your machine and reads/writes files on your machine. We didn't have to build a desktop wrapper, sandbox, or update mechanism — Claude Code already solves all of that.

The trade-offs we accepted

Building on Claude Code means we can't sell Scout to people who don't already use Claude Code. That's a real cost. Here's what we get in exchange:

• No tokens markup. If Scout were a SaaS, we'd have to charge a markup on Anthropic API costs to cover infrastructure. Because you bring your own Claude Code, our pricing only covers Scout itself — nothing pays for someone else's GPUs.
• No model lock-in to us. If we vanish tomorrow, Claude Code keeps working and so do your conversations, transcripts, and skills. The work you've done with Scout doesn't evaporate.
• You get the latest model on day one. When Anthropic ships Claude 5, Claude Code picks it up and Scout immediately runs on it — no migration, no waiting for us to plumb a new SDK.
• Audits are free. Every Scout run produces a Claude Code transcript you can inspect or share. If a message looks wrong, you can see exactly which prompt produced it.

What it means for your data

Because Scout runs inside Claude Code on your machine:

• Your leads stay on your laptop. Lead lists, scoring rationale, message drafts, reply history — all of it lives in local files Scout writes to your home directory. Nothing is uploaded to a Scout server.
• Anthropic sees your prompts (like any LLM call). When Claude Code calls Anthropic to reason about a lead, the prompt text leaves your machine and lands at Anthropic. That's the same trust boundary as using Claude Code for any other task. Read Anthropic's privacy policy to decide what you're comfortable with.
• We don't see anything. The only network traffic Scout itself generates is anonymized analytics (page views, downloads) and license validation. Your leads, ICP, and outreach never touch our servers.

What it means for your wallet

Scout's pricing only covers Scout. You pay Anthropic separately for Claude Code:

• Claude Pro ($20/mo) — generous quota for individual use; works fine for Scout's typical workload of a few dozen leads/day.
• Claude Max ($100–200/mo) — higher quota for power users running Scout aggressively or in parallel sessions.
• Anthropic API credits — pay-per-token, best for teams or programmatic use cases.

This is on top of Scout's free tier or Pro/Team subscription. Yes, it's two bills. The math still works out cheaper than Apollo + Outreach + Lemlist combined for the same scale.

Getting started

If you don't have Claude Code yet:

• Install it from claude.com/claude-code — Mac, Windows, Linux supported.
• Sign in with your Anthropic account, or paste an API key.
• Verify it's working with claude --version in your terminal.

Then come back here and grab Scout. The Scout installer registers Scout's agent skills with your Claude Code install — you'll be running real prospecting flows within minutes.