Clay Expands Claygent Auto-mode, SCIM, Audiences, and MCP Controls

In October and early November 2026, Clay introduced a series of updates to its platform for automating B2B sales and go-to-market processes. The changes covered AI research on companies and contacts, user management, audience synchronization, buyer signal processing, and the security of connections through Model Context Protocol (MCP).

On October 27, the company introduced Auto-mode for Claygent, a mechanism that automatically selects a language model for each research task based on its complexity and cost. Updates released on October 19 and November 2 added automated user management through SCIM, updates to Audiences records using external identifiers, the transfer of signals to Workflows, credit spending limits, control over approved MCP clients, and personal data deletion tools. These changes expand Clay's capabilities as a platform for continuous CRM enrichment, target account management, and marketing operations automation.

Claygent Auto-mode Selects an AI Model for Each Row

On October 27, Clay launched Auto-mode, a new operating mode for Claygent, its AI agent for researching companies, contacts, and other information from public sources. Previously, when configuring Claygent, users selected one language model for an entire table column. Simple and complex tasks were therefore handled by the same model, even when their requirements differed significantly.

For example, determining whether a well-known company is a B2B provider may require a single search. Researching a less familiar company's product portfolio and management structure, however, may require several sources and more complex analysis. Auto-mode selects a model separately for each row. The system uses less resource-intensive models for simple tasks and more capable models for complex ones. Users no longer need to choose a single model for all records.

The mode is enabled by default for new Claygent configurations in tables, Use AI columns, Claygent Builder, and the corresponding Workflow nodes. Existing Claygent configurations continue to use their previously selected models; their settings are not changed automatically.

How Claygent Optimizes AI Research Costs

Clay worked with Applied Compute to develop Auto-mode. The companies used a set of representative research tasks to evaluate answer quality across different models. These results were used for additional training of a router that predicts which model is best suited to a given task. Selection takes into account expected answer quality and execution cost.

Clay also established minimum quality requirements for specific research categories. The aim is to prevent the system from saving credits at the expense of consistently worse results for certain tasks, such as finding contact information.

In tests on real working queries, Auto-mode reduced costs by 47% compared with using the highest-performing model for every task, while maintaining comparable quality. However, this figure comes from Clay's and Applied Compute's own tests. It does not mean every customer will reduce costs by 47%; savings depend on query complexity, the companies being researched, and how tasks are distributed among models.

Clay does not charge a separate fee for Auto-mode. Costs depend on the model that actually performs each task. The company recommends testing the mode on a sample of 5–10 rows before a large-scale run to assess quality and credit consumption.

Workflow Credit Limits Control Automation Spending

On November 2, Clay added Workflow Credit Limits, a mechanism for limiting the number of credits an individual Workflow can consume. Similar limits previously existed for Workbooks. Administrators can now assign a budget directly to an automated process.

For example, a company may have a Workflow that retrieves new CRM leads daily, researches companies, finds contacts, and enriches records. Every operation may consume Clay credits. If the number of new records rises sharply, the automation could consume substantially more resources than planned. Workflow Credit Limits makes it possible to set an upper spending limit.

This is particularly useful for processes that run continuously without manual initiation. Budgets can be controlled separately for different Workflows based on priority and expected data volume. A credit limit controls spending, not the accuracy of results: it does not verify whether contacts or AI research findings are correct.

SCIM Provisioning Automates Employee Access Management

On October 19, Clay introduced SCIM Provisioning for enterprise workspaces. SCIM (System for Cross-domain Identity Management) synchronizes employee accounts between a corporate identity provider and Clay. Supported identity providers include Okta and Microsoft Entra ID, formerly Azure Active Directory.

To configure it, an administrator opens Settings → Workspace → SCIM provisioning and sets up the integration through the WorkOS Admin Portal. Corporate identity provider groups can then be mapped to Clay roles. Four roles are supported: Admin, Editor, Viewer, and Sales Rep.

When an employee is added to the appropriate group, their account is automatically created in Clay with the assigned role. If they leave the company or lose access to the group, the corresponding permissions in Clay are automatically revoked. This reduces manual administration and the risk of former employees retaining access.

Prior Single Sign-On configuration is not required to use SCIM. The feature is particularly useful for agencies and B2B companies whose teams share data and automation infrastructure.

Write to Audiences Updates Contacts Using External Record IDs

In its October update, Clay changed the Write to Audiences node. A Workflow can now find an existing record by its External Record ID and perform an Upsert operation. Upsert combines two actions: updating an existing record or creating a new one if the identifier is not found.

For example, a company transfers Salesforce contacts to Clay Audiences, each with its own CRM ID. When synchronizing again, Clay can use that identifier to update the existing record instead of creating a duplicate. This is especially important when several automations modify data simultaneously. Previously, concurrent runs could create duplicate records.

According to Clay, matching by External Record ID helps prevent these duplicates, including cases involving concurrent execution of multiple Workflows. For ABM, it helps maintain more consistent company and contact data. Changes in job titles, lead status, or segment membership can be sent to Audiences while preserving the connection with the original CRM record.

Synchronization accuracy depends on the stability of external identifiers. If different systems assign incompatible IDs to the same entity, separate matching logic will be required.

Signals to Workflows Connects Buyer Signals With Automation

Clay also simplified transferring Signals results to Workflows. Signals identify events that may matter to marketing and sales teams, such as personnel changes, changes in target account characteristics, and other available signal types.

Previously, additional steps could be required between reviewing signal results and using them in an automation. Users can now open a preview of Signals results in Audiences and send them directly to a Workflow trigger, reducing intermediate actions.

For example, a team identifies companies matching a chosen signal. Those results can be passed to a Workflow that conducts further company research, checks alignment with the Ideal Customer Profile, and prepares data for subsequent use. The mechanism connects event discovery with additional information processing.

This does not mean that all signals automatically trigger any action without additional configuration. Workflow logic, permissions for its operations, and spending limits must be configured separately.

MCP Client Allowlisting Controls External AI Connections

On November 2, Clay introduced MCP Client Allowlisting. The feature allows administrators to restrict which AI clients can connect to a workspace through MCP. Clay uses Model Context Protocol to provide external AI applications with access to selected platform functions.

For example, an employee may connect Clay to Claude or ChatGPT and run permitted research operations directly from an AI assistant. A corporate team may use several different MCP clients, making external connections harder to oversee without a centralized policy.

The new mechanism allows administrators to define approved clients by domain. Connections that do not meet the policy can be blocked. This restricts the use of unapproved AI tools and maintains control over access to corporate data.

MCP Client Allowlisting complements Clay's existing controls for user permissions, access to Functions, and credit consumption.

Claude and ChatGPT Access Without Full Platform Permissions

Clay is also developing a separate MCP access model for sales employees. Administrators can assign a Sales Rep role, giving employees access to approved Clay functions through an AI client without full access to the platform's web interface.

For example, a Revenue Operations team creates a Workflow for checking a company's fit with the Ideal Customer Profile and enriching contact information. After the associated Function is published, a salesperson can run it through Claude or ChatGPT. The salesperson cannot arbitrarily modify tables, inspect every workflow, or manage API keys.

An administrator specifies the available Functions and a monthly credit limit. Individual restrictions can also be applied to particular employees. When the limit is reached, new operations are blocked until the next billing period or an administrator changes the limit.

Enterprise customers have additional controls over access to Audiences. For example, queries can be restricted to accounts assigned to a particular salesperson in Salesforce. These capabilities existed before November's Client Allowlisting; the new update adds another control layer by restricting the MCP clients themselves.

Audiences Privacy Deletion Removes Personal Data Through an API

The November 2 release introduced Audiences Privacy Deletion, intended to process requests to remove personal data from Audiences. Clay added two public API endpoints.

• POST /audiences/privacy-delete accepts a person identifier, such as an email address or LinkedIn slug, and starts a task to delete associated records.
• GET /audiences/privacy-delete/{job_id} checks the status of that task.

Deletion is processed asynchronously. If a task temporarily fails, retries are supported. Clay also reports cascading deletion of associated data and tracking of the operation's status. Task status information is retained for up to five days after completion or failure.

The feature allows organizations to automate handling requests for personal data deletion. For example, if a contact asks a company to remove their information from processing systems, the company can initiate the relevant operation through the API.

Deleting records from Clay Audiences does not automatically delete data from CRM systems, advertising platforms, or every external data provider. Privacy compliance requires considering the entire infrastructure where personal information is stored.

Sculptor Handoff Transfers Enrichment Searches to an AI Assistant

Another November change concerns Enrichment Search. Clay added an option to hand a research task directly to Sculptor, its built-in AI tool for creating and configuring data operations.

For example, a user searches for an existing Enrichment to verify a particular company attribute. If no suitable tool is found, the user can select Use Sculptor. The AI assistant receives the original search query and helps configure the required operation. This is especially useful for nonstandard tasks that are not fully covered by the catalog of ready-made integrations.

A marketing team might look for a way to verify whether a company uses a particular technology on its website or to extract information from several public sources. If a standard Enrichment is insufficient, Sculptor may suggest an alternative configuration.

Sculptor's capabilities are not unlimited. According to Clay's documentation, it is best suited to AI columns, formulas, and individual enrichment operations. It does not guarantee the automatic creation of an entire complex Workflow from a single request.

Metro Area Search Finds Companies and People Across Metropolitan Areas

On October 19, Clay expanded geographic search filters. Metro Area Search allows searches by metropolitan area rather than individual city. The database includes 3,056 metropolitan areas in 151 countries.

The filter applies to several data types: a person's place of residence, the location of company offices, and the geographic locations of previous jobs. For example, a B2B company can search for prospects in the New York metropolitan area without manually compiling a list of all its cities and municipalities. The same approach can be used to find professionals in major business centers across Europe or Asia.

Some Clay tools, including GTM Agent, Sculptor, and Clay Plugin/CLI, automatically account for geographic filters. This simplifies geographic segmentation of accounts and contacts. However, Metro Area Search relies on available location data and does not confirm that a person is physically present in a region at the time of the search.

Clay Ads Reporting and Audiences Source Management Updates

On October 12, Clay introduced changes to the Audiences interface. Sources and export destinations are now combined in the All People and All Companies views. Users can see which systems send data into Audiences and where the data is exported. There is also a more direct path to creating a Workflow for exporting a selected segment.

This is useful for companies that combine CRM data, enrichment providers, and advertising platforms in one system. Clay Ads reporting was also updated. The platform now displays advertising audience sizes and Match Rate according to the definitions used by individual advertising systems.

For example, People Sent indicates the number of records sent, while Audience Size represents the size of the audience created by the advertising platform itself. Match Rate is used as provided by the platform; when it is unavailable, Clay calculates the figure independently. This helps align Clay reports with advertising account data.

Match Rate is not a measure of the accuracy of company or individual identification. It describes the ability to match submitted identifiers to users on an advertising platform and does not confirm that the original record identifies the correct entity.

Using the Clay Updates for ABM and Lead Generation

For B2B marketing, a significant combination of these updates is the automation of the process from signal detection to action. For example, a company creates a target account list using CRM data, buying signals, and Ideal Customer Profile characteristics. Clay Audiences serves as the base of companies and contacts, while Signals detects events requiring additional research.

A Workflow receives the results and runs Claygent Auto-mode to verify information. The processed data is returned to Audiences with an Upsert based on the external identifier. The marketing team can then build segments and send them to available activation channels. Sales employees access approved Functions through Claude or ChatGPT, while administrators control credit spending, user permissions, and the list of permitted MCP clients.

This architecture reduces manual data transfers between systems. Automation does not eliminate errors in source data. For example, if an enrichment provider incorrectly associates a contact with a company, subsequent Workflows may propagate that error into the CRM and advertising segments. Deduplication, account matching checks, contact data quality controls, and the ability to exclude incorrect records are therefore necessary.

Availability of the Clay Updates

Clay's official changelog associates the features with several releases:

• October 12, 2026 — Audiences Sources & Exports and Clay Ads reporting accuracy.
• October 19, 2026 — SCIM Provisioning, Audiences Upsert, Signals to Workflows, and Metro Area Search.
• October 27, 2026 — Claygent Auto-mode.
• November 2, 2026 — Workflow Credit Limits, MCP Client Allowlisting, Audiences Privacy Deletion, and Sculptor Handoff.

Some features in Clay's catalog are marked Beta. The availability of particular capabilities may depend on the subscription plan, workspace configuration, and current rollout stage. Audiences features and additional enterprise data controls, in particular, may be restricted to Enterprise plans. When evaluating implementation, it is therefore important to check not only whether a feature appears in the changelog, but also whether it is available in the account in use.

Clay Continues Developing Its Go-to-Market Automation Platform

Clay is a platform for finding companies and contacts, enriching B2B data, segmenting audiences, and automating sales and marketing processes. Its October and November updates develop the platform in several directions.

Claygent Auto-mode selects an AI model for each research task and reduces the cost of large-scale data enrichment. SCIM and MCP Client Allowlisting expand enterprise control over users and external AI connections. Audiences Upsert and Signals to Workflows help combine data and automate actions based on events. Workflow Credit Limits adds spending control, while Privacy Deletion simplifies personal data removal requests.

For ABM teams, these capabilities support more consistent processes for account discovery, contact research, and transferring data into CRM and advertising channels. Sales teams can use prepared GTM automations through familiar AI interfaces. However, the results depend on source quality, accurate identifiers, CRM consistency, and the reliability of the research operations themselves.

Sources

1. Clay — Auto-mode for Claygent
2. Clay — Product Roundup: October 12, 2026
3. Clay — Product Roundup: October 26, 2026
4. Clay — Product Roundup: October 5, 2026
5. Clay University — MCP Settings
6. Clay University — Audiences
7. Clay University — Sculptor