Cursor
Harriet provisions your organization's skills and connectors straight into Cursor, so developers get the same company knowledge and tools in their editor that they have everywhere else. Cursor sits alongside your other surfaces rather than replacing one.
How it works
Cursor reads the same two open formats Harriet already produces, so nothing has to be re-authored:
- Skills land in Cursor's personal skills folder. People invoke them by typing
/in the Agent chat, and Cursor's agent also picks them up on its own when a task matches. - Connectors land in Cursor's personal MCP configuration, each one routed through Harriet's proxy so tool calls use your organization's credentials and are audit-logged like every other surface.
The cursor-agent command-line tool reads both of these too, so the terminal agent picks up the same skills and connectors with no extra setup.
Cursor is additive. Turning it on does not replace a Harriet Desktop or Claude Desktop option — a machine can run Cursor alongside any of them.
Set it up
- Enable the target. Turn on Cursor in Settings, or pick it during onboarding.
- Install the Harriet Endpoint agent on each machine and sign in. This is the same background agent the Claude Desktop surfaces use; if a machine already runs it, there is nothing new to install.
- Restart Cursor once the device has checked in. Cursor reads its connector configuration at startup, so connectors provisioned while it is running do not appear until it restarts. This is the step people most often miss.
After the restart, open the Agent chat and type / to see your organization's skills, and check the MCP section of Cursor's settings to see your connectors.
What Harriet manages — and what it leaves alone
Harriet only ever writes Cursor's personal configuration in your home folder. It never writes configuration inside a project folder, because connector entries contain a credential that must not end up committed to a shared repository.
- Your own connectors and skills are preserved. Anything you added to Cursor yourself stays exactly as it is; Harriet merges its entries alongside them.
- Removing a device removes only what Harriet added. See removing a device.
Using Harriet as Cursor's model provider
This part is optional, and unlike the other surfaces it is a manual step. Cursor keeps its API key and model base URL inside its own application database rather than a configuration file, so Harriet cannot set them for you.
If you want Cursor's model calls to run through Harriet — using your organization's BYOK keys or managed credits, with usage analytics and budgets applying — copy the OpenAI base URL, API key, and model names from either place:
- the device in Endpoint AI, using the Connect Cursor to the Harriet gateway panel, or
- Harriet Endpoint on the machine: under Agents, open Cursor's ⋮ menu and choose LLM gateway URL.
Paste those values under Cursor → Settings → Models. Once you send a message, the console panel shows Connected.
Uncheck Cursor under Agents in Harriet Endpoint if you do not want Harriet to apply skills and connectors on this machine. That choice stays on the device; it does not turn Cursor off for the rest of the organization.
Two limits worth knowing. Cursor's Tab and inline-edit features always use Cursor's own backend and cannot be routed through Harriet. And because Cursor speaks the OpenAI format for custom providers, your organization needs a billable OpenAI route before models appear in the panel.