Merge for Workforce

Your employees are already using AI at work. Merge for Workforce decides what that AI can reach: which models, which company systems, and under whose permissions. You configure it once from one dashboard. Employees keep the tools they have (Claude, ChatGPT, Cursor, Claude Code) and point them at Merge, which checks every request against your identity provider, scans it for sensitive data, and writes it to a log.

Use it when the people using AI work for you. When the AI user is your customer, you want Agent Handler.

Merge pushes approved models, MCPs, and skills into the tools employees already run, so the sanctioned path is also the one that needs no setup.

Pick the capabilities you need

Capability	What you govern	Start here if
AI Gateway	Which models each team can use, how requests route, and what they cost	Model spend is the problem you have this quarter
Connectors	Which company systems an employee’s AI can touch, and what data those calls carry	AI needs to reach Slack, Jira, Salesforce, or Drive, scoped per team
Skills and agents	The procedures and MCP servers employee agents load, and a record of what they ran	Employees are running agents and you want them following your playbooks

All three read the same identity, the same Groups, and the same dashboard. Adding the second capability is a configuration change, not another rollout. Run AI Gateway and Connectors together and a single log covers both halves of a request: the model that answered, and the system it touched.

What you configure

Surface	What it does	Start here
Identity	Connects Okta, Azure AD, or any SCIM-compatible provider, so employees arrive and leave with your directory	SCIM provisioning
Tool access	Assigns Tool Packs to Groups, with overrides for the people a Group does not fit. Thousands of maintained tools sit behind it	Managing tool access
Guardrails	Reads tool call payloads against entity and regex rules, then flags or blocks on a match	DLP rules
Models	Sets the models each team can use and routes their traffic through AI Gateway	AI Gateway
Skills	Publishes your procedures so agents run them instead of inventing their own	Publishing skills
Devices	Ships a light desktop client through your MDM to configure machines and enforce policy locally	Workforce desktop client
Logs	Keeps tool call logs and the Audit Trail in the dashboard, and streams both to your SIEM	Tool Call Logs

Configure only the rows for the capabilities you adopted. Most teams begin with identity plus either AI Gateway or tool access, then add the desktop client when they want device-level inventory and enforcement.

How your employees connect

Tools and models are separate connections. An employee can have one, the other, or both.

To tools

One MCP endpoint, one SSO sign-in. A consent screen lists the tools the employee’s Group grants, and the client renews its own token after that. Connecting tools over MCP covers the setup: SCIM, Group mapping, and the endpoint itself.

Ask for something outside that set and the call fails, then surfaces in your dashboard as an access request. Approve it, decline it, or widen the Group.

To models

Point the AI tool at AI Gateway instead of the model provider: the Gateway base URL, and a Gateway API key where the provider key used to go. In Claude Code that means ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN. Every harness has an equivalent, and Coding agents and IDEs works through them one at a time. For direct API use, start with Get started.

From there your routing policy chooses the model and the team’s budget caps what it can spend.

Or let the desktop client do both

The desktop client writes both connections on the employee’s machine: MCP configuration for tools, a Gateway base URL for models, and the skills you publish. Turn on enforcement and it blocks the direct provider endpoints too, which leaves the governed path as the only one that works. The employee configures nothing. See Workforce desktop client.

Either way, access is bound to identity. Deprovision someone in your identity provider and their tokens go with the account, so AI access ends when the rest of their access does.

What it is not

Not a customer-facing agent platform. If the end user is your customer, read Building an agent.
Not an AI client. Employees keep Claude, ChatGPT, Cursor, and Claude Code. Workforce governs what those clients can reach.

Next steps

Connect tools over MCP

Set up SCIM, map Groups to tools, hand employees the MCP endpoint

Workforce desktop client

Deploy through your MDM, inventory the AI already installed, enforce policy on the device

Guardrails

Catch sensitive data in a tool call before it reaches a third-party system

AI Gateway

Set the models each team can use and see spend by team, tool, and model