How to Run Your GTM Stack With Claude Code: 7 Real Workflows

The mental model: the agent drives, the software does the work

The mistake people make with AI in sales is asking the agent to replace the tools. That is backwards. Your CRM, your enrichment platform, and your sequencer are good at things an agent will never be good at: storing state reliably, sending mail from warmed domains, holding the system of record. An agent like Claude Code is good at the opposite: reading across all of those tools at once, reasoning about what to do, and driving them through a multi-step task without you clicking between tabs.

The wire between the two is MCP. When a tool ships an official MCP server, Claude Code (or Codex, or any MCP client) can read and write it directly. You connect Attio, Clay, and Fireflies.ai once, then ask in plain language and the agent does the clicking.

Here is the division of labor, then seven workflows that put it to work.

What each side does best

Rule of thumb: if the task is "hold this reliably" or "send this safely," that is the software. If the task is "look at all of this and decide," that is the agent.

1. Pipeline hygiene in one prompt

Goal: stop deals rotting silently.

The software (your CRM) holds every deal, its stage, and the last-touch date. Claude Code is what reads all of that, spots the pattern, and acts. Ask it: "List deals in Attio over $10k that have had no activity in 14 days, and for each draft a short re-engagement email referencing the last thing you discussed." The CRM could never write the email. Claude could never be the durable record. Together you get a triaged list with drafts attached, in one pass.

2. The pre-call brief

Goal: walk into every call prepared without 30 minutes of prep.

Before a meeting, Claude reads the account in your CRM, pulls fresh firmographic and signal data from Clay, and checks the last call transcript in Fireflies.ai. It returns a one-page brief: who you are meeting, what was promised last time, what changed at the company since, and three questions to ask. Clay does the enrichment it is built for. Claude does the synthesis no single tool can, because the inputs live in three different products.

3. Post-call follow-up and CRM update

Goal: never lose a commitment made on a call.

This is the highest-leverage one. After a call, the transcript lands in Fireflies.ai. Ask Claude to read it and: draft the follow-up email with the agreed next steps, update the deal stage and key fields in your CRM, and create tasks for each commitment. The recorder captures, the CRM stores, and the agent does the connective work that usually gets skipped because it is tedious. You review the draft and send.

4. Inbound lead triage and routing

Goal: respond to good leads in minutes, not hours.

A form fills. Claude enriches the contact through Clay, scores it against your ICP, writes a first-touch reply in your voice, and either routes it to the right rep or drops it into the nurture track. Clay supplies the data. Your CRM owns the routing rules and the record. Claude makes the judgment call on fit and drafts the human-sounding reply, which is the part automation platforms do badly.

5. List building with real personalization

Goal: outbound that does not read like a mail merge.

Clay is unmatched at building and enriching a list through waterfall data. What it cannot do is write a genuinely specific opening line for each prospect. Claude can: feed it the enriched rows and ask for a one-line personalized angle per contact, grounded in something real from the data. Then Instantly or lemlist sends from warmed domains. Three tools, each doing the one thing it is best at, with the agent writing the part that decides whether the email gets a reply.

6. The weekly pipeline review, written for you

Goal: a forecast narrative without the Sunday-night spreadsheet session.

Your forecasting tool, Clari for example, holds the numbers. Claude queries the pipeline, compares it to last week, and writes the narrative: what moved, which deals slipped, where the risk is, and what to chase first. Dashboards show you the numbers. They do not tell you the story or what to do about it. That is the agent's job, and it can do it every Monday at 8am on its own.

7. Sequence drafting and reply analysis

Goal: better cold email, faster, backed by your own data.

Instantly and lemlist handle sending and hold every reply. Claude reads the reply data, tells you which subject lines and angles actually got responses, and drafts the next round of variants to test. The platform owns deliverability and the send. The agent owns the analysis and the copy. This closes the loop that most teams run on gut feel.

How to set this up

1. Pick tools that ship an official MCP server. Most of the leaders in the AI-readiness report do, including Attio, Close, Clay, Fireflies.ai, and Outreach. A tool with no MCP and no open API cannot be driven this way.
2. Connect each MCP server to Claude Code (or your agent of choice) once. After that you prompt in plain language.
3. Start read-only. Let the agent draft and summarize before you give it write access to your CRM. Trust grows from there.

The one rule that keeps this safe

Keep a human on anything that leaves your company. Let Claude draft the email, update the field, and build the list. You approve what gets sent. The agent is fast and tireless and occasionally confidently wrong, so the split is simple: it does the work, you own the send.

For the data on which tools are actually built for this, see the 2026 AI-readiness report. For the wider picture, read how to build an AI-native sales stack.