What Is Zoho MCP and How to Use It Right
In Integrating AI (Claude) with Zoho CRM, Optimus showed what AI can do when it connects directly to Zoho. This post covers the connection itself in detail: how many ways there are, how to set each one up, where the limits are, and how to get AI to answer accurately while using tokens efficiently.
I. WHAT IS ZOHO MCP
MCP (Model Context Protocol) is an open standard that lets AI use external tools. With MCP, Claude or ChatGPT can read and update data directly in your apps, with no need to export a file and paste it into the chat.
Zoho MCP is Zoho's ready-made implementation of this standard, and it currently supports 69 Zoho apps.
Tools are the specific actions AI is allowed to take in an app, such as finding a customer, creating a deal or updating a ticket. Each MCP server provides a set of tools. AI reads each tool's description to pick the one that fits the question.
How it works
Because it acts through the user's own account, AI can only do what that account is allowed to do in Zoho.
II. OPTION 1: OFFICIAL CONNECTORS BUILT BY ZOHO
In Claude
Go to Customize › Connectors and search for "zoho".
Pick the connector labeled "by Zoho" and click +.
Sign in to your Zoho account and grant access.
Claude currently has Zoho-built connectors for these apps:
Zoho CRM, Books, Desk, Projects, Show, Bigin, Vani and Catalyst.
Zoho Analytics: available only in the Claude desktop app.
The list also includes a Zoho CRM connector built by a third party, so make sure you pick the "by Zoho" one.
In ChatGPT
Open the Plugins tab and search for "Zoho CRM" (Developer: Zoho).
Click + and sign in to Zoho.
This is currently the only Zoho plugin in ChatGPT. Whether you can install it depends on your plan and workspace settings, so some accounts may not see it yet.
Limits
Zoho picks the tools for you. The Zoho CRM connector has 41 tools, including high-risk ones such as deleting records, editing user details and creating fields.
Claude lets you block individual tools. Tools are split into two groups: read-only and write/delete. Each tool can be set to one of three levels: Always allow, Needs approval, or Blocked.
One connector per app. If you use both CRM and Desk, you have to connect twice.
Note: Claude puts the data query tool (executeCOQLQuery) in the write/delete group. If you block that whole group, AI also loses its most efficient query tool. Block write/delete tools one by one instead of blocking the whole group.
III. OPTION 2: ZOHO'S PRE-CONFIGURED SERVERS
Zoho provides many ready-made MCP servers, each built for a type of work. Every server already comes with a set of tools, so you just create it and connect.
Recommended steps:
Go to mcp.zoho.com and click Create MCP Server.
Pick a server under Pre-configured servers.
The list currently includes servers for Zoho CRM, Desk, Books, Mail, Projects, WorkDrive, Bigin, Cliq and a few other apps. For Zoho CRM alone, there are these servers:
CRM Data & Metadata Operations
CRM Activities & Engagement
Lead Management System
Deal Lifecycle Tracker
The server is created right in your Zoho account, and you can still add or remove tools afterwards.
Once it's created, copy the Server URL and add it to Claude:
In Claude, go to Customize › Connectors and click Add › Add custom connector.
Give it a name, paste the URL into the Remote MCP Server URL field, and click Add.
Select the connector you just added, click Connect and sign in to Zoho.
Servers listed in the Zoho CRM help docs. Zoho CRM also publishes 4 servers grouped by how much they can change:
Data Insights: read-only.
Data Operations: create, read, update, delete.
Module Customization: edit modules, fields and layouts.
Workflow & Process Automation.
The URLs on that page belong to the India data center (ending in .zohomcp.in). If your Zoho account is in another data center, create the server from Pre-configured servers as described above, so the server sits in the same data center as your data.
Pros:
Get started quickly with toolsets Zoho has already grouped by type of work.
Easy to limit permissions, for example by using only a read-only server.
Optimus uses the Zoho CRM Data Operations server for its own internal work.
Cons: each server covers one type of work in one app. If you need several apps in one connection, use option 3.
IV. OPTION 3: BUILD YOUR OWN SERVER ON ZOHO MCP
This is the most flexible option, and the only one that lets several apps share a single connection.
Step 1: go to mcp.zoho.com, click Create MCP Server, then choose Create MCP server to create an empty server.
Step 2: click Add Tools, pick an app, then pick the tools you need.
Step 3: on the Connection tab, choose how access is granted.
Step 4: copy the Server URL, add it to Claude as a custom connector as in option 2, then sign in.
Pros:
One server can serve several apps, for example CRM and Desk for a customer service team.
Turn individual tools on or off right on the server, and the change applies to every AI that connects to it.
Logs record every time AI calls a tool.
Things to watch:
The URL contains an API key, so keep it secure.
Use Authorization on Demand. If everyone shares one account, every action is recorded under one person's name and everyone has the same permissions.
Claude Free can only add 1 custom connector. In ChatGPT, you need a Business plan or higher for full use; the Pro plan can only read data.
V. COMPARING THE 3 OPTIONS
Option 1: Official connector | Option 2: Zoho's pre-configured server | Option 3: Your own server on Zoho MCP | |
Setup | Find it in the list, click connect | Pick a pre-configured server, paste the URL | Create a server, pick tools, paste the URL |
Choosing tools | Zoho picks them; Claude lets you block individual tools | Zoho picks them by type of work; you can add or remove | You pick each tool |
One connection for several apps | No | No | Yes |
Best when | You want to start quickly | You want a lean toolset for one type of work, such as read-only | You need tight control, or the app has no connector yet |
VI. COST
Zoho MCP has no separate fee. Every time AI calls Zoho, it counts against your company's existing API quota.
This quota is shared across the whole company, so when the whole team starts using AI, keep an eye on usage for the first few weeks.
VII. AFTER CONNECTING: AN EXAMPLE OF FINDING DEALS AT RISK OF SLIPPING
Optimus tested this on a sample CRM with 10 deals to review that week, and asked Claude: "Which deals are at risk of slipping this week? Why?"
Without instructions | With instructions | |
Response time | about 40 seconds | about 30 seconds |
Data AI had to read | about 24,000 tokens | about 3,000 tokens |
Result | included unrelated deals | exactly the 10 deals to review |
Response time depends on how many deals match the filter, not on the total number of deals in the CRM. AI filters inside Zoho, and Zoho returns results in under half a second. Most of the wait is AI reading and analyzing the matching deals. The more deals match, the more AI has to read and the longer it takes to answer.
Without instructions, AI guessed that the closed stage is called "Closed Won". The company in this example doesn't use that name, so AI also pulled in old, unrelated deals.
With instructions that spell out which deals are in scope and what the sales stages are called, AI filtered correctly on the first try, read about 7 times less data and answered faster.
Best practices
Write instructions for AI.
Spell out which app and module to use for which task, the names of your sales stages, and the filters you use most.
You can put them in the Instructions section of a Claude project.
The more modules your company has, the more this matters: Optimus's CRM has 116 modules, and Invoices, Sales Orders and Purchase Orders each have two modules with the same name.
Enable only the tools you need.
Safer: AI can't delete data if the delete tool isn't enabled.
Fewer tokens: each tool's description also takes up room in AI's working memory, and some tools are very heavy.
Better tool choice: the more tools there are, the easier it is for AI to pick the wrong one.
Start with read-only access. Keep create, update and delete tools on Needs approval, and loosen them gradually once the team trusts the answers.
Ask with specific filters. Zoho returns data in under half a second. Most of the wait is AI reading and processing the data, so the more specific the question, the faster the answer.
CONCLUSION
How you connect determines how much control you have: official connectors to start quickly, pre-configured servers when you want a lean toolset for one type of work, and your own server when you need to pick each tool and share one connection across several apps. To get accurate answers, though, the most important part is the instructions you give AI.
FREQUENTLY ASKED QUESTIONS
Which option should a team of 5-10 people choose?
If the whole team only uses CRM: the official connector is enough, with the delete and user-editing tools blocked.
If the team uses several apps, or everyone needs the same set of rules: build a server on Zoho MCP with Authorization on Demand, so an administrator manages the tools in one place.
Should you use Claude or ChatGPT with Zoho?
Claude has more Zoho-built connectors (CRM, Books, Desk, Projects and a few other apps). Even the Free plan can add 1 custom connector to try it out.
ChatGPT only has the Zoho CRM plugin so far, and needs a Business plan or higher to make full use of your own server.
Is your Zoho data used to train AI?
That depends on the AI provider's policy, not on Zoho. By default, Anthropic and OpenAI don't use data from business plans for training; personal plans follow different policies.
Check this before letting staff use personal accounts with customer data.
How is Zoho MCP different from Zia?
Zia is the AI built into the Zoho interface. Zoho MCP lets outside AI such as Claude or ChatGPT work with your Zoho data. You can use both side by side.
Optimus is a Zoho Authorized Partner in Vietnam. Contact contact@optimus.vn for advice on and implementation of AI integrations with Zoho.
Comments