Cursor is an AI code editor, so connecting it to Zoho MCP suits developers and technical consultants more than everyday business users. Once connected, Cursor's agent can read and, if you allow it, change Zoho data while you work on code, for example looking up CRM field names while writing a Deluge script or an API integration. Zoho lists Cursor among its supported MCP clients and supports both automatic and manual configuration. For creating the server and choosing its tools, follow Zoho MCP Implementation Guide.
The setup in outline
- In the Zoho MCP console for your data centre, create a server and add only the tools you need. UK and Irish customers on the EU data centre use the EU console.
- Open the server's connection options, go to the Cursor tab under MCP Clients, and copy the JSON configuration Zoho provides. It already contains your server's URL.
- Paste that configuration into Cursor's
mcp.json file. Zoho's instructions use the user-level file at ~/.cursor/mcp.json. You can reach the same setting through Cursor's settings, under Tools and MCPs.
- Open the chat and set it to Agent mode. Zoho's tools are available there, and not in the other modes.
- Try a harmless read-only request first, such as listing a few records.
The one thing that catches developers out
The configuration file holds a secret. Zoho says the MCP URL contains a secure API key. That key now sits, in plain text, in a file on your computer. Anyone who can read the file can use your server, with whatever tools you added.
- Keep it out of repositories. If you use a project-level configuration file instead of the user-level one, make sure it is excluded from version control. A key committed to a shared or public repository should be treated as leaked.
- Never paste the configuration into a chat, ticket, or document when asking for help.
- If it leaks, generate a new URL. Zoho MCP lets you do this, but you then have to reconfigure each client with the new one.
- One server per person or purpose makes it easier to switch off access without disturbing everyone else. Servers can also be disabled from the Zoho MCP console.
Using it sensibly
| Do |
Why |
| Use a sandbox or test organisation while building | An agent that writes code and calls live tools in the same session can change real data by mistake. |
| Expose read-only tools first | Looking up metadata and example records is most of the value for development work. |
| Check Cursor's tool approval settings | Look for the option that asks for confirmation before a tool runs, and keep it on for anything that writes data. We have not described the exact setting because Cursor changes its menus. |
| Review the generated code before running it | Check any Deluge or API code the agent writes, especially anything that deletes or updates in bulk. |
The general safety advice in our safe setup checklist applies to Cursor as much as to any other assistant.
How it differs from ChatGPT and Claude
- It is set up per computer, through a file, rather than per account through a web page. Each developer configures their own copy.
- The secret lives on disk, so protecting the machine and the repository matters more than it does with a browser-based connector.
- It is built for developers. Choose it when the goal is to build and test Zoho integrations or Deluge. For business users who want to ask questions of their data, a chat assistant is a better fit. See Connecting ChatGPT to Zoho MCP for that route, and Using AI with Zoho for how we choose.
If it does not work
- No Zoho tools appear. Check the chat is in Agent mode, then restart Cursor so it rereads the configuration file.
- It connected, then stopped. The server may have been disabled, or the URL regenerated, in the Zoho MCP console. Paste in the new configuration.
- A JSON error. Make sure you pasted the configuration inside the existing structure of the file, not on top of other servers already listed there.