How to Set Up Claude Code for GTM Outbound
Install to first working workflow in one afternoon. The exact steps.
What You Need Before You Start
Three accounts and one machine. An Anthropic Claude account (Pro tier at $20/mo is the starting point). An account with whichever CRM your team uses (HubSpot or Salesforce). An account with your enrichment vendor (Clay, Apollo, or similar). A laptop or desktop running macOS, Linux, or Windows with WSL.
The whole setup runs in a terminal. If you've never used a command line before, plan an extra hour to get oriented. If you've used Git or run a Python script, you have what you need.
Step 1: Install Claude Code (15 minutes)
On macOS or Linux:
curl -fsSL https://claude.ai/code/install.sh | bash
On Windows, install through WSL (Windows Subsystem for Linux) using the same command. Direct Windows install also works through the Anthropic installer but most GTM Engineers run WSL for the Unix-like environment.
After install, run:
claude
This launches the interactive Claude Code session in your terminal. The first time, it walks you through authentication. Use your Claude Pro subscription credentials. Within two minutes you should be in a working session.
Step 2: Set Up Your Project Directory (10 minutes)
Create a project for your GTM workflows. Most teams use a directory like ~/gtm-agents/ or a Git repo at your team's GitHub org.
mkdir ~/gtm-agents && cd ~/gtm-agents && git init
Inside this directory, you'll add a CLAUDE.md file (project context), a .mcp.json file (shared MCP server config), and subdirectories for each workflow. Keep this simple. Don't over-architect on day one.
Step 3: Wire the CRM MCP Server (20 minutes)
For HubSpot:
claude mcp add --transport http hubspot https://mcp.hubspot.com/anthropic
For Salesforce, you'll either install a community MCP server or write a custom wrapper. The full setup is documented in Claude Code MCP Salesforce.
After adding the server, run /mcp inside a Claude Code session to authenticate. For HubSpot, this opens a browser window for OAuth. You log in, approve the scopes, and the token gets stored.
Start with read-only scopes for the first workflow. You can add write scopes when you build the workflows that need them.
Step 4: Wire the Enrichment MCP Server (20 minutes)
For Clay:
claude mcp add --transport http clay https://api.clay.com/mcp --header "Authorization: Bearer YOUR_CLAY_TOKEN"
Store your Clay token in an environment variable rather than hardcoding it. Use your shell's standard pattern (export CLAY_TOKEN in .zshrc or .bashrc).
For direct enrichment vendors (Apollo, FullEnrich, Cognism), wrap their REST APIs in a small stdio MCP server. The Anthropic SDK includes a template that takes about 40 lines of Python. See the enrichment workflow guide for the full pattern.
Step 5: Write Your CLAUDE.md (30 minutes)
This is the most important file. It's the persistent context every Claude Code run loads. Three sections.
About Us. One paragraph on what your company does, what you sell, who you sell to. Keep it tight.
Our ICP. The explicit firmographic rules. Industries, employee size, geography, funding stage, tech stack signals. Bullet list, not prose. The model uses this for scoring and qualification.
Voice and Style. Four to six lines on tone, sentence length, words to avoid, what a good first touch looks like, with two example emails that hit the bar.
Don't over-engineer this on day one. A tight 50-line CLAUDE.md is better than a sprawling 500-line one. You'll refine it over weeks as you see what the agent gets wrong.
Step 6: Build the First Workflow (45 minutes)
Start with account research. This is the safest first workflow because it's read-mostly and produces immediately useful output.
Open a Claude Code session in your project directory:
cd ~/gtm-agents && claude
Prompt the agent: "I want to build a workflow that reads a list of company domains from a Clay table called 'research_queue' and writes a research brief to each company's HubSpot record. Help me write the workflow as a runbook in research-accounts/RUNBOOK.md."
The agent walks you through the structure. You iterate together. After 30 to 45 minutes, you have a working RUNBOOK.md that you can execute with claude -p "process the research queue".
Step 7: Test on 10 Accounts (30 minutes)
Add 10 real company domains to the Clay table. Run the workflow.
Open each enriched account in HubSpot by hand. Verify the research brief is accurate. Verify the right fields got updated. Verify nothing else got touched.
When you find a miss, fix the prompt or the RUNBOOK. Don't hand-edit the bad record. Rerun. Recheck. Repeat until the 10-account sample is clean.
Step 8: Schedule for Production (20 minutes)
Once the workflow produces clean output reliably, schedule it. The simplest deploy is a cron job on your laptop (only useful if your laptop is always on) or on a small cloud VM ($5 to $10/mo).
On a Linux VM, edit the crontab:
0 7 * * 1-5 cd ~/gtm-agents && claude -p "process the research queue" >> logs/research.log 2>&1
This runs the workflow every weekday morning at 7 AM, logging output to a file. Add a Slack alert on failure using a separate cron line that checks the log for errors.
What to Build Next
Once account research is stable and your team is using the output, add one more workflow per week. Suggested order:
Week 2: Pre-call brief generator (read-only, low risk).
Week 3: Enrichment and scoring workflow with CRM write-back (moderate risk).
Week 4: Reply triage classifier (higher risk, lots of trust-building first).
By month two, you should have four working workflows that save your team 15 to 25 hours per week. That's the goal for a first deployment. Don't try to ship 12 workflows in a quarter. Ship four well, learn from the team's reactions, then expand.
For the next-level patterns, see AI SDR with Claude Code, the sales agent guide, and the full use case list.
Authoritative References
For the install and configuration commands, see Anthropic's Claude Code documentation. For MCP server setup, see the Claude Code MCP documentation.
Frequently Asked Questions
How long does it take to set up Claude Code for GTM outbound from scratch?
One afternoon from install to first working workflow if you're already comfortable in a terminal. Hour one: install Claude Code, set up the Anthropic API key, run the first interactive session. Hour two: wire HubSpot or Salesforce via MCP, wire Clay or your enrichment vendor. Hour three: write the first CLAUDE.md and the first prompt for an account research workflow. Hour four: test on 10 accounts and tune. Headless cron setup and the higher-stakes workflows come on day two or three.
Do I need an Anthropic API key separate from my Claude Pro subscription?
Depends on which tier of Claude Code you use. Claude Pro ($20/mo) includes Claude Code with usage limits suited to most individual GTM Engineers. If you want headless cron runs without rate limits or pay-per-token billing, you can also use the Anthropic API directly, which requires a separate API key from console.anthropic.com. Most teams start with the Pro subscription and add API access only when they need higher-volume headless jobs.
What's the minimum MCP server stack for a GTM outbound workflow?
Three servers cover most outbound workflows. Your CRM (HubSpot or Salesforce) for the source of truth on accounts and contacts. Your enrichment platform (Clay or direct vendors like Apollo) for firmographic and contact data. Your sequencer (Smartlead, Instantly, or Lemlist) for staging outbound. Add Slack for alerts and a database MCP (Postgres or your warehouse) if you have one. That's the working set for 90% of GTM outbound workflows.
Should I install Claude Code on a server or my laptop?
Laptop for interactive development and testing. Server (a small cloud VM) for scheduled headless runs. Most GTM Engineers do both. Install Claude Code on the laptop for the build and test phase. Once a workflow is stable, also install it on a small DigitalOcean or Hetzner VM (usually $5 to $10/mo) and schedule it as a cron job there. This way the workflow keeps running when you close your laptop.
What's the first GTM outbound workflow I should build with Claude Code?
Account research. It saves real time (20 to 40 minutes per account), reads from CRM and enrichment without writing anything sensitive, and produces output your SDRs immediately use. Start there. Get one workflow polished and trusted. Then build the second one (usually enrichment-plus-scoring or pre-call brief generation). Don't try to ship five workflows at once. Ship one well, learn what your team needs, then expand.
Source: State of GTM Engineering Report 2026 (n=228). Salary data combines survey responses from 228 GTM Engineers across 32 countries with analysis of 3,342 job postings.