Your reps already work in ChatGPT. Your prospect data lives in Clay. Every time a rep stops to research an account, they leave the chat for a dozen tabs and lose the thread, so a lot of the research just never happens.
Clay's MCP connector closes that gap by bringing Clay into ChatGPT itself. A rep can find a contact, verify the email, research the account, and draft the outreach in one conversation, without ever opening Clay. RevOps can also package its best workflows so reps run them with a single prompt. This is how to connect it and use it well.
What you need before you start:
- A ChatGPT account (browser or desktop) or Claude, and a Clay account: You can create the Clay account during setup and get 500 free credits to start.
- For the team rollout, a Clay workspace admin: To set credit limits and package workflows reps run by name.
- That is the whole prerequisite: The point of MCP is that reps never have to open Clay.
Step 1: Connect Clay to ChatGPT (or Claude)
Setup happens inside the AI tool and takes about a minute.
In ChatGPT, type @Clay at the start of a prompt in the browser, or /Clay in the desktop app, then describe what you need. The first time, you authenticate through app.clay.com/oauth; sign in or create an account and you get 500 free credits to start. In Claude, add the Clay connector from the Claude Connectors page (on Claude Enterprise an admin enables it first), then just ask Claude to find or research in plain language. MCP, the Model Context Protocol, is the layer doing this: it connects your Clay workspace to AI tools like ChatGPT, Claude, and Codex, so the same data and workflows show up wherever the rep already works.
Auto-plays a prompt routing through MCP to Clay and back. Click a node to hold it and read what it does.
Your prompt
You type: "@Clay find the VP of Sales at stripe.com and verify their email"
MCP puts Clay's data and workflows inside the AI tool the rep already uses, so research-to-action happens in one conversation instead of a tool switch.
Step 2: Find and verify contacts inside a prompt
The workflow is find, filter, enrich, in that order, one prompt at a time.
Ask for the people you want, narrow the list, then enrich the ones that survive. Run it as three short prompts rather than one overloaded request, because each step lets you check the result before spending the next credit. Clay returns an interactive view you can toggle to a table, filter, and enrich with new data points (verified email, work history, recent activity). One rule saves most of the headaches: search by company domain, not name, so "stripe.com" returns Stripe and not a same-named company.
1) @Clay find marketing leaders at stripe.com2) Filter to US-based directors and above3) Enrich the remaining contacts with a verified work email
Step 3: Research the account before you reach out
Good context is the difference between a reply and an archive, and now it lives in the same chat.
Before writing anything, ask Clay to research the account: hiring trends, tech stack, funding history, recent product launches, and leadership changes. Two habits make the results clean: use the company's domain instead of its name, and research one company at a time so the AI does not blur two accounts together. The research the rep pulls here carries into the next step automatically, so the outreach is built on what the account actually cares about, not a generic template.
Auto-cycles the four rep workflows with example prompts. Click one to hold it and read its prompt and output.
Example prompt
@Clay find sales leaders at stripe.com
Inside the chat a rep can do the four jobs that used to span a dozen tabs: find and verify contacts, research the account, write the outreach, and prep the meeting, with context carrying between them.
Step 4: Draft personalized outreach in the same conversation
Writing outreach used to be a multi-tab scavenger hunt. Now it is one more line in the chat.
Because the people search and account research already happened in this conversation, the context carries over: the AI knows who you are writing to and what you found. Two setup steps make the drafts usable. First, confirm your business context in your Clay workspace Settings so the AI writes as your company, not in a vacuum. Second, enrich the contact's email before you ask for a draft, because the "Draft email" action only appears once there is a verified address to send to. Then ask for the message and refine it in the same thread.
Step 5: Give reps your best workflows as MCP Functions
Reps get the most out of the connector when they run the workflows your ops team already built.
A Function is a reusable Clay workflow that an admin builds once and turns on with Enable for MCP, giving it an actionable name and description. After that, reps invoke it from ChatGPT, Claude, or Codex with a single prompt and never open Clay. That is what keeps quality consistent: instead of every rep writing their own research prompt and getting their own result, they all call the same vetted "Company enrichment" or "Account brief" Function and get the same output. RevOps builds the logic; reps execute it where they already work.
Auto-plays ad-hoc prompts vs one packaged Function. Click a lane to hold it and inspect it.
Packaging a workflow as an MCP Function turns ad-hoc rep prompting into one vetted workflow every rep calls by name, so output stays consistent and reps never open Clay.
Step 6: Set guardrails, and know when to stay in the Clay platform
An always-on connector to 150+ providers needs limits, and it is not the right tool for every job.
From the MCP section in the Clay side nav, a workspace admin sets credit limits per team member and monitors usage, so no one burns the workspace's credits from a chat window. Then match the tool to the task. Use Clay in ChatGPT for the rep-scale work: 1 to 20 contacts, exploratory research, ad-hoc planning, and individual emails. Stay in the Clay platform for 20+ contacts, complex automated workflows, and deep CRM syncs. The rule of thumb: if a rep is typing the same prompt over and over, that workflow should become a Function or a table, not a habit.
Auto-sweeps task size across the ChatGPT vs Clay-platform threshold. Drag the slider or click Replay to inspect.
Clay in ChatGPT
do it in the prompt- Exploratory research
- Ad-hoc planning
- Individual emails
Repeating a prompt? If you are typing the same prompt over and over, package it as a Function or build it as a table.
Admins set per-rep credit limits and monitor usage from the MCP section of the Clay side nav.
Clay in ChatGPT is for rep-scale, ad-hoc work (1 to 20 contacts); volume, automation, and CRM sync belong in the Clay platform, and any repeated prompt should become a Function.
Used this way, the connector gives reps hours back, because the research and writing happen where they already are.
time Anthropic's lean GTM team saved by automating their Salesforce opportunity upserts in Clay, instead of comparing lists by hand
Read the full storyCommon failure modes, and how to avoid them
Most Clay-in-ChatGPT setups stumble the same few ways. Watch for these.
- Searching by company name, not domain: "Stripe" can match the wrong company; "stripe.com" will not. Use the domain, and research one company at a time.
- Asking for a draft before enriching the email: The draft action only appears once a verified email exists. Enrich the contact first, then ask for the message.
- Letting every rep prompt from scratch: Freehand prompts give inconsistent output. Package the vetted workflow as an MCP Function so every rep runs the same one.
- No credit guardrails: An open connector to 150+ providers can run up credits fast. Set per-rep credit limits and monitor usage from the MCP nav before rollout.
- Using it for the wrong job: A 2,000-contact list or a CRM sync does not belong in a chat window. Keep rep-scale, ad-hoc work in ChatGPT and move volume and automation into the Clay platform.
Outcomes teams report after putting Clay in their reps' hands
What teams report after giving reps Clay-powered research and enrichment
“We consolidated three vendors into Clay and started enriching data points that didn't exist in any traditional database. Our reps went from starting every conversation cold to knowing exactly who to call and what to say.”