Expose an employee over MCP
In this guide, you’ll expose an employee as an MCP server, create an access token and connect to it from your AI tool. You can then ask the employee questions, assign it tasks and follow their progress without leaving Claude or Cursor. It takes about 10 minutes.
Before you start
Section titled “Before you start”- The Builder role or above, and access to the employee you want to expose. This guide uses
Ada. - An Admin who can turn on inbound MCP for the workspace, if it isn’t on yet.
- Claude Desktop, Claude Code or Cursor installed on your computer.
Expose and connect the employee
Section titled “Expose and connect the employee”Turn on inbound MCP
An Admin goes to Settings > Capabilities and turns on Inbound MCP. You do this once per workspace.
Enable Ada's MCP endpoint
Open Employees, select Ada and go to Integrations > MCP. Click Enable MCP endpoint, then choose which tools callers get:
| Tool | What it does | Default |
|---|---|---|
ask |
Asks Ada a question and returns her answer. Runs as a short task with Read and Draft actions only. | On |
assign_task |
Creates a task with a goal, inputs and an optional deadline. Returns a task ID. | On |
get_task_status |
Returns the task’s status, its latest steps and its result when done. | On |
cancel_task |
Stops a task and its current run. | Off |
Click Save. Orqane shows Ada’s endpoint URL:
https://mcp.orqane.com/<workspace>/adaChoose how callers sign in
Under Authentication, keep OAuth on so workspace members can sign in with their Orqane account. Claude Desktop uses this.
For Claude Code and Cursor, click Create token and fill in:
| Field | Value |
|---|---|
| Name | My laptop |
| Scope | Employee Ada |
| Expires | 90 days |
| Allowed tools | ask, assign_task, get_task_status |
Click Create. Copy the token now: Orqane shows it only once.
Connect your AI tool
Add Ada to your tool. Replace <workspace> and <token> with your values.
Open Settings > Connectors and click Add custom connector. Fill in:
| Field | Value |
|---|---|
| Name | Orqane – Ada |
| Remote MCP server URL | https://mcp.orqane.com/<workspace>/ada |
Click Add, then Connect, and sign in with your Orqane account. Ada’s tools appear in the tools menu of the chat box.
Run this command in your terminal:
claude mcp add --transport http orqane-ada \ https://mcp.orqane.com/<workspace>/ada \ --header "Authorization: Bearer <token>"Start Claude Code and run /mcp to check that orqane-ada is connected.
Open ~/.cursor/mcp.json, or .cursor/mcp.json in your project, and add:
{ "mcpServers": { "orqane-ada": { "url": "https://mcp.orqane.com/<workspace>/ada", "headers": { "Authorization": "Bearer <token>" } } }}Open Cursor Settings > MCP and check that orqane-ada shows a green dot.
Try it
In your AI tool, write:
Ask Ada for a summary of what she learned about our competitors last week.Your tool calls ask, and Ada answers from her memory and knowledge. Then try a longer job:
Assign Ada a task to update the competitor table with this month's pricing,then check its status in a few minutes.Your tool calls assign_task, then get_task_status. The run also appears in Ada’s Activity tab, with the client’s name.
You’ve exposed Ada as an MCP server and connected to her from your own AI tool.
What the caller can and can’t do
Section titled “What the caller can and can’t do”- The caller talks to Ada, not to her tools. It can’t call HubSpot, Gmail or any other tool directly. Ada decides which tools to use.
- Autonomy still applies. A task that arrives over MCP is a trigger like any other. At L1 Approve, Ada still waits for her human manager’s approval before she sends an email, and
get_task_statusreturnswaiting_for_approvalmeanwhile. - Policies still apply. Workspace policies and guardrails run on every step, whoever started the task.
- Costs go to Ada. Each call is charged to Ada’s budget and shows in Costs.
- Everything is audited. The audit log records which member or token made each call.