Connect Cursor to MintMCP
Connect Cursor to MCP servers managed by MintMCP, and extend the same governance to Cursor's headless Cloud Agents.
Cursor connects to MintMCP in three ways, depending on how you run it:
| Path | Mode | How it authenticates |
|---|---|---|
| Direct MCP connection | Desktop | Interactive OAuth |
| Agent Monitor hooks | Desktop | Organization key |
| Agent identity | Cloud Agents | Bearer token or OAuth client credentials, or per-user OAuth via the Cursor dashboard |
For desktop use, the first two paths work together: the MCP connection governs which tools the agent can call, while the hooks give you visibility into everything the agent does, including non-MCP activity like shell commands and file reads.
Connect Cursor to a Virtual MCP
Add a Virtual MCP (VMCP) to Cursor's MCP config for governed tool access.
- In MintMCP, open the VMCP and copy the MCP URL from the MCP Client Setup tab
- Add the server to
~/.cursor/mcp.json(or via Cursor Settings → MCP):
{
"mcpServers": {
"mintmcp": {
"url": "https://app.mintmcp.com/s/abc123/mcp"
}
}
}
Some organizations use org-scoped URLs (https://app.mintmcp.com/o/<org>/s/<slug>/mcp), so copy the exact URL from the VMCP's MCP Client Setup tab rather than constructing it.
- Sign in through the OAuth flow when Cursor prompts
Once connected, the VMCP's tools appear to Cursor's agent, and every tool call is logged and governed by MintMCP.
Monitor Cursor with Agent Monitor
Cursor runs its own local tools (shell commands, file reads, edits) that the MCP connection does not see, so MintMCP provides hook scripts that report this activity to Agent Monitor. Rules configured in MintMCP (secrets detection, tool permissions, org rules) can then allow, block, or prompt for approval on prompts, shell commands, MCP calls, and file reads.
Follow Set up Agent Monitor for Cursor to generate an organization key and install the hooks.
Connect Cursor Cloud Agents
Cursor's Cloud Agents (formerly called Background Agents) run in isolated cloud VMs without the desktop IDE open. They clone a repository, work on a branch, and push changes, typically as pull requests, and can be triggered from the Cursor desktop app, the web UI, Slack, GitHub comments, or the Cursor CLI.
How Cloud Agents handle MCP
MCP configuration works differently from the desktop IDE:
- MCP servers are configured through the Cursor web surfaces, not
.cursor/mcp.json: personal servers through the MCP dropdown at cursor.com/agents, and team-shared servers through Dashboard > Integrations & MCP (admins). The project-level config file is not picked up by Cloud Agents. - HTTP and stdio transports are supported (SSE and
mcp-remoteare not). Cursor recommends HTTP, because tool calls proxy through Cursor's backend and credentials never enter the agent's VM. - OAuth is supported per user: the user authorizes the MCP server once through the Cursor web dashboard, even for team-shared servers.
- Bearer token and API key auth also work, as a static token in an
Authorizationheader or environment variables for stdio servers.
You can connect a Cloud Agent to a Virtual MCP either way: complete the OAuth flow from the Cursor dashboard for per-user access, or use an agent identity when the agent should act as itself.
Connect through an agent identity
Give the Cloud Agent its own identity when you want its tool calls attributed to the agent rather than a user, with a tool set scoped to its job:
- In MintMCP, create an agent identity for the Cloud Agent workflow, with the connectors and tools the agent needs
- Issue a bearer token (or OAuth client credentials) against the agent identity
- In Cursor's Cloud Agent MCP configuration, add the agent identity's MCP URL with the bearer token in the
Authorizationheader
This gives the Cloud Agent governed tool access through MintMCP, with full audit attribution to the agent identity, scoped permissions, and rule enforcement.
Agent Monitor hooks are not available in Cloud Agent VMs, since the hook system depends on local Cursor configuration files.
Next steps
- Set up Agent Monitor for Cursor for visibility into desktop agent activity
- Agent identities to understand how MintMCP attributes and scopes agent access
- Bearer keys to issue machine-to-machine credentials