Get the Clay + Claude repo to supercharge your GTM
Run the /clay command inside of Claude and access 16 skills across 6 workflow stages, 21 providers, 10 templates and 8 cost guardrails
There’s a take going around that I keep running into that I think it’s ridiculous, and it’s that Claude is about to replace Clay. They say “You won’t need the enrichment tool anymore because the model can find the data itself.”
I don’t buy it, and I’ve been running both every day to prove it to myself.
Clay and Claude do different jobs. Clay is the engine: more than a hundred data sources, the waterfalls that try the cheap source before the expensive one, the credit economics, the orchestration that turns a raw list into a clean, scored, enriched table. Claude Code is the operator that drives the engine. Asking a model to replace Clay is like asking it to replace your CRM. Wrong layer.
I don’t hear enough people talking about how to use them together.
I run Clay across a handful of SaaS client workspaces, and I got tired of rebuilding the same tables and re-explaining the same rules every time I opened a new one.
So I packaged the whole thing into a workbench that Claude Code can drive. It’s public and free. Below I’ll walk through what’s in it, the one command that runs it, and a few plays you can steal this week.
One quick note before we get into it. To follow along you’ll need a paid Clay workspace, the official Clay MCP connected, and Claude Code (or Cursor). That’s the whole setup.
What’s in the Clay Repo
It’s a system: 16 sub-skills plus a /clay router that reads what you’re trying to do and sends you to the right one. They’re grouped by the six stages every Clay workflow moves through.
Hygiene – clay-list-clean pre-cleans a raw source list (dedup, domain and title normalization, suppression) before you spend a single credit. – clay-data-hygiene keeps your CRM deduped and refreshed on a schedule instead of rotting between quarters.
Build – clay-abm-list builds account-keyed target lists with the trigger events baked in. – clay-account-research and clay-prospect-research run multi-source briefs on a company or a person (priorities, news, hiring, leadership, a warm-path hypothesis). – clay-enrich-waterfall handles the email, phone, and LinkedIn waterfalls.
Score – clay-icp-score lays in a 10-point fit composite. – clay-buying-signals fuses intent from eight-plus sources with time decay, and it won’t go live until the signal beats a real lift test against your closed-won.
Activate – clay-outbound writes personalized copy behind a Sending Gate and pushes it to your sequencer. – clay-inbound-routing wires a webhook to enrich, route, and Slack the right rep. – clay-signal-monitor watches a trigger and routes what it catches.
Quality and cost – clay-claygent-iterator tests a Claygent prompt before you scale it. – clay-credits forecasts spend and compares provider mixes. – clay-cost-audit runs a portfolio-wide waste check across every workbook. – clay-troubleshoot fixes the broken or expensive ones.
Persistence – clay-template-library lets you save, version, and reuse your best workbooks instead of rebuilding them from memory.
Every one of these enforces the same habits (gates before credits, free sources before paid), so the cheap moves are the default and not the thing you remember to do at the end.
Try this one command
Here’s what running it looks like. I open Claude Code in the workbench folder and type one thing:
/clay build a Tier 1 ABM list of US Series B fintechs that hired a new VP of Sales in the last 90 days. Score them against my ICP and draft outbound for the top tier.
From there the router takes over. It reads this as an account build plus scoring plus activation, and it runs an eight-question intake first so it builds my workbook and not a generic one. Then it cleans the list, drops an ICP gate column before any paid enrichment touches a row, builds the waterfall free-source first and paid only as fallback, lays in the 10-point score, and writes the outbound behind a Sending Gate so nothing exports until it clears my rules.
What used to be an afternoon of clicking is now a command and a review. I’m approving and editing, not building from scratch. And the structure it stands up is the disciplined version, the one I’d build on my best day, instead of the rushed version I’d put together at 4pm with three other things on fire.
Why this beats doing it by hand
The speed is nice. What saves me money is the discipline it forces.
In my experience, cleaning the list first cuts credit burn by something like a third to a half, because you’re not paying to enrich duplicates and dead rows. Gates before credits means I never spend on a record that was never going to qualify. And before I scale any Claygent column, the iterator runs it on 5 rows, then 25, then 100, and scores each batch for pass rate, fabrication rate, and credits per row. That’s how I catch a made-up data point at row 25 instead of finding it at row 5,000 after I’ve already paid for all of them.
It won’t read your mind, though. The intake is the work, and thin answers build thin workbooks.
Plays you can run this week
A few things people are already doing with it:
Rebuild your best-performing workbook as a template, then spin it up for a new segment or a new client in minutes instead of an afternoon.
Stand up an inbound demo router: form fill, enrich, ICP score, and a Slack ping to the right AE in close to real time.
Point clay-signal-monitor at a trigger you care about (fresh funding, a key exec hire, a tech install showing up) and let it route hot accounts straight into your cadence.
Generate one-page account briefs for your AEs the morning before a big meeting.
Run clay-cost-audit across every workbook once a month and cut the ones that aren’t earning their credits.
Clean a messy CSV before it ever touches a paid column (the cheapest habit on this list, and the one most people skip).
Most of the plays here aren’t original to me. I built on open-source Clay skill packs from forma-norden, mariosworkflows, and the ColdIQ crew, then wired them into one system with the cost guardrails I run for clients.
Do this today (it will take 15 min)
If you want to try it today:
Clone the repo into your skills folder: git clone https://github.com/guerrilla2799/clay-workbench ~/.claude/claudecodeskills/clay-workbench
Connect the official Clay MCP in your connector settings.
Run /clay against one real goal you have this week, not a test list.
Start small. Let it build a single list, then check the ICP gate column before you spend a credit.
It runs MCP-first, but if you’d rather drive Clay yourself the first time, it falls back to a manual UI walkthrough and tells you what to click.
It’s public and free at github.com/guerrilla2799/clay-workbench. Clone it, run one command against a real list, and send me what you build. I want to see it.
And if this is the kind of AI-in-the-GTM-stack workflow you want more of, I break one down every week in Stack & Scale.
– Brandon