Zoho MCP: How to Connect Zoho to Claude and ChatGPT Safely

What is Zoho MCP?

MCP, the Model Context Protocol, is an open standard that lets an AI client call tools on another system. Zoho MCP is Zoho's hosted product for building MCP servers. Each server holds a set of tools, and each tool performs one specific action, such as searching CRM records, creating a Books invoice or replying to a Desk ticket.

Zoho describes MCP as an execution layer, not a data store, and says it does not depend on any one language model. The AI client decides which tool to call; Zoho runs the call with the permissions of the account that authorised it.

• Zoho apps listed on the Zoho MCP page include Zoho CRM, Bigin, Books, Billing, Expense, Desk, Mail, Cliq, Creator, People and Survey, with Projects, WorkDrive and Calendar mentioned in its FAQ.
• Some apps also document their own MCP servers, including Zoho Analytics, Zoho Assist, Zoho Lens, Zoho Projects and, in India, Zoho Payments.
• Third-party services such as GitHub, Notion, Stripe and Twilio can be added to the same server.

How do you set up a Zoho MCP server?

The steps below follow Zoho's implementation guide. Labels can change, so confirm them in your console.

1. 1Sign in to the Zoho MCP console with the Zoho account whose data the assistant should reach.
2. 2Click Create MCP Server and name it by purpose, for example Sales read-only. Zoho also offers pre-configured servers such as CRM Data & Metadata Operations.
3. 3Open Tools, click Add Tools, filter by service and add only the tools this use case needs.
4. 4Under Connections, choose Authorize on Demand (each user signs in) or Authorize via Connections (a shared connection). Third-party services use connections only.
5. 5Copy the server URL from the Connect section. It carries a key, so treat it like a password.
6. 6Add it to your client and test with a read-only prompt before enabling any write tools.
7. 7Check the Logs section to confirm which tools ran.

Zoho's guide suggests keeping a server to about 300 tools or fewer so the client chooses tools reliably. In practice, a server of 10 to 30 tools per job is easier to control and audit.

How do you connect Zoho MCP to Claude and ChatGPT?

After you save, the client redirects you to a Zoho MCP authorisation screen. Click Allow, accept the permissions and the tools appear in the client. Anthropic notes that custom connectors are not verified by it, so connect only servers you control and review each tool approval, especially write actions.

How are Zia Agents different from Zoho MCP?

Zoho MCP lets an outside assistant use Zoho. Zia Agents are AI agents that run inside Zoho. Zoho announced Zia Agents in February 2025 with prebuilt agents such as an SDR agent, a customer support agent and an account manager agent, plus Agent Studio and an Agent Marketplace for sharing agents.

In Zia Agent Studio you describe a task or configure an agent yourself: instructions, knowledge base documents, tool groups and guardrails written as Do's and Don'ts, such as discount limits. Zoho says you can also connect external models such as ChatGPT and Gemini with your own keys. Choose MCP when your team already works in Claude or ChatGPT; choose Zia Agents when the work should stay inside Zoho apps for every user.

Either way, agents are only as good as the data and permissions underneath, which is why this work often starts with a review by a Zoho CRM implementation partner before any tool is switched on.

What should you check on security, permissions and audit?

• Permissions: tools act with the rights of the authorising user. Authorise with a user whose Zoho CRM profile and role match the job, not a super admin.
• Authorisation mode: Authorize on Demand keeps each person on their own access. A shared connection gives every client user the same rights.
• Server URL: anyone with it can call the tools. Regenerate the key if it leaks and update clients.
• Logs: Zoho MCP logs every tool call with status, tool, user ZUID and execution ID, kept for 30 days. Zoho CRM's own audit log records data changes for about 60 days.
• Data flow: Zoho says MCP does not store responses, but data returned to the AI client is then handled under that client's terms, so check your plan's data and retention settings.
• Revoking: disable or delete the server, revoke user authorisation, or remove the connector from the client.

Under GDPR, PIPEDA or HIPAA, treat the AI client as a processor of any personal data it receives and record that in your data processing records.

What mistakes do teams make with Zoho MCP?

• Adding every tool for an app in one server, including delete actions.
• Testing write tools on live data instead of a Zoho CRM sandbox.
• Connecting before the CRM is clean. An assistant reading duplicate or stale records gives confident but wrong answers.
• No owner for the server, so nobody reviews logs or removes leavers.

Clean data and clear profiles matter more than the AI client you choose, so fix duplicates and access rules before any tool is switched on.

How does Flowpeak help with Zoho MCP?

Flowpeak is a Zoho partner based in Chennai, India, delivering remotely worldwide. We design MCP servers per use case, set profiles and authorisation modes, test in a sandbox and document what each tool can change. MCP and agent projects are quoted in writing after a free scoping call. To plan your first safe use case, Book a free call.

Frequently asked questions