
Connect your Clay workspace to AI tools — and manage rep access, credit limits, and Functions from one place.
MCP (Model Context Protocol) is how Clay connects your workspace to AI tools like Claude, ChatGPT, Microsoft Copilot, and Glean. Workspace admins control which reps have access, how many credits they can spend, and which Functions are available to them — all from Settings → MCP users.
Getting started
The minimum setup to get reps running is an invite and a credit budget. Functions and Audiences are additive — they improve the experience but aren't required to start.
- Go to
Settings → Teamand invite each rep. Assign them the Sales Rep permission type to give them MCP-only access — they won't be able to see the Clay UI, tables, API keys, or workflows. - Go to
Settings → MCP usersand clickSet default limitto set a workspace-wide credit budget. This applies automatically to every rep who connects, so set it before reps start. - Enable any Functions you want reps to invoke from their AI tool (see Enabling a function for MCP below).
- (Enterprise only) If your workspace uses Clay Audiences, configure the Audiences controls to give reps access to first-party CRM data.
- Share the relevant connection guide with your reps:
Enabling a function for MCP
Functions are reusable enrichment workflows built in Clay that reps can invoke directly from their AI tool with a single prompt. Admins build them once and enable them for MCP — reps never need to open Clay to use them.
- Go to the
Functionstab in your workspace and find the function you want (or click+ New functionto create a new one). - Click the function to open its settings and toggle
Enable for MCP.- Set a name and description for the function — this is what reps see when browsing available functions, so make it actionable (e.g., "Company enrichment waterfall" or "Outbound email generator").
- Confirm the function has an enrichment or output column. Toggling MCP and setting inputs alone won't return data without a column that produces results.
- Avoid function names that overlap with Clay's built-in tools (e.g., "find contacts" or "enrich company") — the AI tool may invoke the default instead.
For more information about building functions, see Functions.
Setting credit limits
Credit limits cap how many Clay credits a rep can spend through their AI tool in a given month. Credit spend resets on the 1st of each month at midnight UTC.
There are two levels of control:
- Default credit limit — applies automatically to all new MCP users when they first connect. Click
Set default limitto configure. Reps without an individual override inherit this limit. - Per-user override — find the rep in the user table and click the pencil icon next to their
Credit limitto set an individual amount. Their current usage tracks against this limit in real time (e.g.,0 / 1,000). Reps showingNo limithave no cap applied.
Monitoring usage
The MCP users table gives a live view of every rep who has connected Clay to an external platform:
- Name — rep's name and email address
- Platforms — icons indicating which platforms the rep has connected (Claude, ChatGPT, etc.)
- Credit limit — the rep's current limit, either the workspace default or a per-user override
- Credits used — live usage tracked against the rep's limit
- Salesforce ID — populated automatically when
Sync user IDs from audiencesis enabled
Use the search bar at the top of the table to find a specific rep by name or email.
MCP credit usage also appears in the main credit usage dashboard at Settings → Credit Usage, alongside all other workspace credit consumption.
Audiences controls
If your workspace uses Clay Audiences, two additional workspace-level toggles appear on the MCP users page:
- Sync user IDs from audiences — continuously syncs audience data to match MCP users to the Salesforce accounts they own. Updates run incrementally every 15 minutes, with a full sync once a week.
- Allow querying all accounts — when enabled, reps can query any account in the synced audience, not just accounts they own in Salesforce.
FAQ
Can admins remove a rep's access to Clay in Claude or ChatGPT?
Yes, by removing them from your workspace. While admins cannot directly revoke a rep's MCP connection from the MCP users page, they can remove the rep from the workspace entirely — if a rep isn't in your workspace, they have no access to your data or workflows. To limit usage without removing access, set their credit limit to 0, which hard-blocks further actions without removing them from the workspace.
When do credits reset?
Credit spend resets on the 1st of each month at midnight UTC.
What happens when a rep hits their credit limit?
Further actions through their AI tool are hard-blocked until the monthly reset — the rep won't be able to run enrichments or invoke Functions. Admins can increase the per-user limit at any time from the MCP users table to restore access immediately.
Where else can I see MCP credit usage?
MCP usage appears in the main credit usage dashboard at Settings → Credit Usage, which tracks all credit consumption across your workspace broken down by table, integration, and time period.
What's the difference between the default credit limit and a per-user override?
The default limit is a workspace-wide setting that applies automatically to any new rep who connects. A per-user override replaces the default for a specific rep. Reps showing No limit have neither a default nor an override applied.
If a rep isn't on Audiences, what value do they get from MCP?
People and company search plus Functions. Without Audiences, reps can still find contacts at target accounts, research companies, and invoke any Functions your ops team has built — including Functions that pull first-party data from Salesforce or Snowflake directly. Audiences adds the ability to query owned accounts and opportunity data from the rep's CRM.
When a rep runs an action through Claude or ChatGPT, does it count as a Clay action?
Yes. Credits work the same way via MCP as they do in Clay directly — there's no MCP surcharge. If a Function that finds someone's email and phone number costs 12 credits in a Clay table, it costs 12 credits when a rep triggers it from Claude or ChatGPT.
Does Clay's MCP use our data to train AI models?
No. Both Anthropic (Claude) and OpenAI (ChatGPT) exclude commercial and API customer data from model training by default. Clay's MCP operates through their commercial APIs, so your workspace data is never used for training. Clay itself does not store customer table data for other customers and acts as a data processor under standard data protection agreements.
How do Anthropic and OpenAI handle data from MCP requests?
- Anthropic (Claude): Commercial and API data is excluded from model training. Data may be retained briefly for abuse monitoring only. For qualifying organizations, Zero Data Retention (ZDR) means data isn't stored at rest after the API response is returned.
- OpenAI (ChatGPT): API, Enterprise, and Business data is not used for training by default. Inputs and outputs may be retained up to 30 days to provide the service and detect abuse, then deleted unless legally required. ZDR is available for eligible endpoints.
For the most current information, refer to Anthropic's privacy policy and OpenAI's usage policies.
What can the AI tool access in our workspace when reps connect via MCP?
The AI tool can only call Clay tools and Functions that a workspace admin has explicitly enabled — it cannot browse tables, view raw workspace data, or access anything outside of what enabled Functions return. Specifically:
- Functions: only those with
Enable for MCPtoggled on - Audiences data: only if admin has enabled
Sync user IDs from audiences - Account scope: admins can restrict reps to only accounts they own in Salesforce
- Credit controls: admins set spending limits per user
How do user permissions work in MCP?
Admins can assign the Sales Rep permission type to restrict users to MCP-only access, which blocks them from the full Clay web app. Reps with this role can only run ops-approved Functions from their AI tool — they cannot view, edit, or create tables directly in Clay.
Troubleshooting
For a full guide to common issues — including connection errors, Function configuration problems, and platform-specific ChatGPT issues — see MCP — Troubleshooting & FAQ.
Explore other docs
Other popular resources
Join a cohort, learn Clay fast!
The faster way to master Clay. Sign in if you're enrolled in a cohort (current or past) or apply!
Hire GTME Talent
Find and connect with GTM talent who've demonstrated expertise in building advanced workflows



