Your first GTM play in Claude Code: a working walkthrough
Skip the theory. Here is a revenue play you can build in an afternoon from your terminal: the config, the commands, and what lands in your CRM when it runs.
Most "GTM engineering" content stops at the architecture diagram. This one doesn't. By the end of this walkthrough you'll have a play that sources twenty companies, researches each one, writes the first-touch email from your positioning doc, and logs everything to your CRM. The tool doing it is the one already open on your laptop.
What a play actually is
A play is a repeatable revenue workflow your agent runs: a trigger, a set of steps, and a destination. "When a new signup matches our ICP, research them, draft an email, and log it" is a play. The trigger can be a schedule, a webhook, or you typing a command. The steps are your agent's tools. The destination is your CRM and your outbox.
That's the whole abstraction. If you can describe the workflow in a sentence, you can build it as a play.
Step 1: connect the MCP server
One entry in your MCP config connects your agent to sourcing, outbound, calls, and content. No per-tool keys, no glue scripts between them.
The server exposes tools like source-companies, draft-outreach, and log-activity. Your agent discovers them the same way it discovers any MCP tool: list the server, read the descriptions, call what fits the job.
Step 2: describe the play in plain English
In Claude Code, describe the workflow: "Source 20 seed-stage fintech companies in the US founded after 2023, not in our CRM. For each, pull the founder, draft a two-sentence first-touch email referencing something specific about the company, and save all of it as drafts for my review."
The agent plans the tool calls, runs them in order, and shows you its work. Review the drafts before anything sends. The first run is where you tune: the ICP filter, the email voice, how much research is enough.
Step 3: save it as a play
Once the run looks right, save the workflow. Committed as code or stored in chat, the play becomes a thing you can rerun, schedule, or hand to a webhook. Next week you change one line instead of re-explaining the whole workflow.
What lands in your CRM
Every run writes one resolved record per company: the sourced firmographics, the research summary, the drafted email, and the send status, attached to the right contact. Nothing lives in a chat scrollback or a spreadsheet export. When a reply comes in, the history is already there.
Where teams get stuck
- Skipping the positioning doc. The agent writes generic email without it. Ten lines about who you serve and why you win is enough input to change the output completely.
- Starting with volume. Twenty well-researched contacts teach you more than two thousand scraped ones. Scale after the play converts, not before.
- No review gate. Keep a human on the send button until the play has run a few cycles. The point of the play is removing the busywork, not removing the judgment.
Time cost, honestly
First play: an afternoon, most of it spent writing the positioning doc and reviewing drafts. Second play: an hour, because the config and the review loop already exist. By the third you're composing plays the way you compose functions, and the marginal cost of a new motion is close to zero.
The GTM engineering guide covers the broader picture; this walkthrough is the on-ramp. RevPeak's MCP server connects with one command, and plans start at $34 a month.