Learn and master Clay with Clay university

This lesson covers custom functions via MCP: what they are, how to build them, and how to enable them so reps can call your workflows directly from ChatGPT, Claude, or other AI tools.

🔧 What Are Custom Functions?

Custom functions are pre-built workflows that live in Clay and get surfaced in your AI tool of choice.

Reps don't need to know how the workflow works or open Clay. They just ask for what they need in their AI tool, and the function runs for them. If they want to dig deeper, they can check it out in the functions library.

Each function you build and enable for MCP is another track available for every platform your reps are using.

📚 Building on Clay Functions

If you've taken the Clay Functions course, you've seen what these workflows look like. You can create a company firmographic enrichment function that pulls revenue, employee count, and monthly web traffic across twelve or thirteen different providers in waterfalls, standardizing the entire workflow across your team.

🚀 Adding the MCP Layer

The MCP layer makes workflows callable from your rep's AI tools—getting functions into ChatGPT, Claude, Glean, and other tools.

For RevOps, this standardizes how your team prospects. Build best practices once and distribute them everywhere. No more reps improvising their own enrichment sequences.

For reps, this means access to sophisticated enrichment flows without building anything.

💬 How Reps Access Functions

Reps access functions in two ways:

1. Browse suggested options - Functions appear as suggested options in chat interfaces.
2. Call them conversationally - Type something like, "Use the LinkedIn to email subroutine to find this person's email," or "Run our company's enrichment workflow on these five accounts." Reference the function by name, and the AI tool handles the rest.

📋 The Six-Step Setup Process for Admins

Step 1: Create the Function

Head to the Functions tab in your workspace and click Create New Function. Give it a clear name.

Step 2: Enable It for MCP

Open the table settings and toggle Enable for MCP. Set a name and description for the MCP app—this is what reps see when browsing available functions, so make it clear and actionable.

Step 3: Configure the Inputs

Click on function inputs, then Edit Source and create whatever inputs your function needs: domain, contact name, company email address, or whatever the workflow requires.

Step 4: Test with Sample Data

Before building out the full workflow, call the function from your tool and ask for a sample record. Say something like, "Use the company enrichment function for test@clay.com." You should see it run immediately, giving you real data flowing through the table, which makes building the rest of the workflow easier.

Step 5: Build the Rest of the Clay Workflow

Add your enrichments, waterfalls, and AI research—whatever you would normally do when building out a Clay table.

Step 6: Configure the Return Data

Click on Send Data Back, then Edit Column, and choose which columns you want returned to the user. This determines what reps see when they run functions. Return only the data that's useful for the rep.

Once you've completed all six steps, the function is live, and reps with MCP access can call it immediately.

✅ Best Practices

1. Name your functions clearly. This makes it easier for reps to call the function using natural language in their chat tool.

2. Start with two or three high-value functions rather than trying to boil the ocean. Get a few right first, then expand.

3. If a function isn't showing up, check that Enable for MCP is toggled on in the Edit Table settings.

4. Enable or disable specific functions from the MCP settings page. This is your governance layer to prevent reps from running expensive or non-standard workflows.

🎯 What's Next

You now know how to take any Clay workflow, enable it in MCP, and surface it inside your rep's AI chat tool.

In the next lesson, we'll switch to the rep side and cover what this looks like when working in ChatGPT or Claude day-to-day.

‍