Freshservice's MCP Gateway lets AI tools read your tickets, assets, and knowledge base. This means less switching between tools, looking up details, and entering the same information again and again. Sounds like a dream for any IT support manager trying to make processes more efficient, right?
Well, that dream comes with two caveats. First, the gateway, still in beta, has a limit of 5,000 calls per month. Second, it works best when the data it accesses is clean and reliable.
If you’re not ready to spend forever cleaning up your entire IT service desk — only to see your call allowance disappear lightning-fast — a phased approach makes more sense. Connect data categories one by one, starting with those that can deliver the most value for the least effort. This is precisely the same approach that applies to AI-ready data migration.
In this guide, we’ll explain what to connect first, what can wait, and why order matters.
What Is the Freshservice MCP Gateway?
Freshworks presented its Gateway at Refresh 2026 in San Mateo on May 14, alongside its Freddy AI Agent Studio. As the company’s CPO, Srini Raghavan, put it: AI tools may be good at reasoning, but they often lack access to an organization’s live service data. Bridging that gap is the problem the MCP Gateway is designed to solve.
Freshworks’ MCP Gateway works in two directions. Inbound MCP allows external AI tools to access Freshservice data, while Outbound MCP lets Freshservice AI Agents take action in connected business platforms.
Freshworks isn’t the only ITSM vendor rolling out MCP connectivity. Their release is part of a broader trend. ServiceNow, for example, has its own MCP – ServiceNow MCP Server Console. So, if you are choosing between these two ITSM platforms, MCP support is what they have in common: both work to make service data more accessible beyond the platform.
Inbound MCP mode: AI tools reading and acting on Freshservice data
Inbound MCP mode gives external AI tools, such as Cursor, Claude, and Microsoft Copilot, access to Freshservice data. A developer working in Cursor can easily check a live ticket’s status without leaving the IDE, while an agent using Claude can review a configuration item’s history before responding to a request.
The endpoint supports more than 30 functions covering tickets, assets, agents and requesters, onboarding and offboarding, the service catalog, and solution articles.
That said, since Inbound MCP mode relies directly on service desk data, this article will focus on preparing for and setting up the Inbound connection rather than Outbound MCP mode.
Outbound MCP mode: Freshservice Freddy AI reading and acting on external data
Outbound MCP mode works in the opposite direction. It enables Freshservice Freddy AI Agents to perform tasks within Atlassian, Notion, Linear, and ClickUp.
For example, a Freddy AI agent handling an onboarding request can create a ticket in Linear and update the related Notion page as part of the same workflow.
Who can use the Freshservice MCP server and how
Both directions remain in Early Access until September 2026. Till then, Inbound MCP mode requires the Enterprise plan, while Outbound MCP mode is available on Growth, Pro, and Enterprise with an active Freddy AI Agent Studio subscription.
During the EAP, Freshworks caps usage at roughly 100 tool calls per minute and 5,000 per month for both. Starting in September, the MCP server will become a full-fledged feature available across Growth, Pro, and Enterprise plans for both modes, with new usage limits:
- Growth — 25 actions per minute; 100 actions per account per month
- Pro — 50 actions per minute; 500 actions per account per month
- Enterprise — 100 actions per minute; 1,000 actions per account per month
Once an account exceeds its monthly allowance, Freshworks charges $15 for an additional 1,000 actions.
These limits are worth factoring in before connecting the gateway to workflows that run frequently. That planning should begin with the data those workflows will rely on.
How AI Amplifies Data Problems
Whether creating a ticket, updating an asset record, or answering a question using the knowledge base, AI agents rely on the available data. If that data contains gaps, duplicates, or outdated information, an agent might carry those problems straight into the next action.
Say your CMDB contains two records for the same laptop. An outbound agent may open a repair ticket against the wrong record while the correct asset history remains elsewhere. If a requester field still points to an employee who has left the company, an onboarding request could be routed to an inactive mailbox, leaving the new hire’s laptop request without a clear owner.
A person reviewing the same records may notice that something looks wrong and pause to investigate. An AI agent might retrieve a record through the MCP gateway and act on it immediately.
So, before asking which tasks the MCP server can help you with, ask what your data is actually ready to support.
Before You Connect Anything: A Five-Point Readiness Checklist
There are five areas where a data problem can quickly turn into an AI agent’s mistake. Check them before granting access. Or, if you are migrating from another service desk to Freshservice, review these items — and the rest of the Freshservice data migration checklist — before moving to the platform.
- Ticket history. Make sure ticket histories are complete. An agent answering “How was this resolved last time?” needs resolution notes to work with. If the original ticket was closed without these notes, there’s little useful context to retrieve.
- Ticket fields. Check that required fields contain accurate values and records aren’t duplicated. Relationships should also hold up: a ticket must point to the correct asset, and that asset to the correct owner.
- Users. Confirm that requester and agent records belong to people who are still active. Otherwise, approvals and assignments risk ending up with a former employee’s account instead of the person responsible today.
- Knowledge base. Solution articles should reflect the current process from beginning to end. Remove or update overlapping articles so an agent doesn’t have to choose between two conflicting answers.
- Current assets. Your inventory should match what is actually in use. Retired, outdated, and duplicate asset records should be cleaned up before an agent starts referencing them.
Cleaning up the data removes one major source of risk, but it doesn’t solve everything. You still need to catch problems early and avoid burning through your usage allowance.
Connecting data categories one by one lets you validate each connection, start training external AI agents before the full rollout is complete, put the MCP gateway to use earlier, and get more value from your call allowance. So, the next question is: what should you connect first?
The Connection Priority Framework: What to Connect First
Every data category behind MCP carries a different amount of context per API call. Tickets return the highest yield for an AI agent. Knowledge base articles come next, and assets follow. Connect the rest only once these three earn their access.
1. Tickets, or the context engine
Tickets carry the richest context per call. A single record often contains the original problem, its context, troubleshooting history, resolution, and timestamp. That gives an agent both the issue and the path to solve it.
Before connecting tickets, make sure their history is complete. This matters especially if you’re also migrating data to Freshservice. CSV imports can leave conversation threads, attachments, or field relationships behind. A dedicated migration tool with field mapping such as Help Desk Migration preserves the full context MCP will expose.
2. Knowledge base articles, or the answer layer
Once tickets are connected, knowledge base articles extend the AI agent’s reach. Instead of piecing together an answer from several related tickets, the agent can retrieve a documented solution in a single call. That means fewer calls and a faster answer for the employee waiting on the other end.
If you’re migrating to Freshservice, Help Desk Migration preserves your entire knowledge base, transferring all its content—including articles, categories, and sections.
3. Assets & CMDB, or the “what broke” layer
Assets answer a different question. Tickets and articles explain what happened and how similar problems were resolved; asset records show what is actually involved.
An agent troubleshooting, say, a laptop issue may need to confirm the model, warranty status, and current owner before deciding whether a ticket history or knowledge base article applies.
4. Users & requesters, or personalization and permissions
User and requester data adds context to what the agent has already learned from tickets, articles, and assets. It can connect a request with the right department, employee, or approval chain.
This layer comes fourth because personalization works best when there is already something concrete to personalize.
5. Service catalog & changes, or connect last
The service catalog and change management data are typically more structured and leave less room for ambiguity. They generally rely on stricter fields and predefined processes, giving an AI agent more predictable information to work with. This makes these categories reasonable candidates for the final stage of the rollout.
As a result, you get the highest-value, most context-heavy data up and running first, then add more structured layers on top.
Setting Up the Connection
As of August 2026, the Freshworks MCP integration remains in Beta. To request access, contact your technical account manager.
Once the Freshservice MCP integration is activated, each agent connects the AI tool they need from their own account.
Authentication options
Freshservice offers two ways to authenticate MCP connections: OAuth 2.0 and an API key. We recommend OAuth 2.0 as the more secure choice.
An API key works like a single password shared across every session. Anyone who gets hold of this key can use it until it’s manually revoked. OAuth 2.0 works differently, issuing a token that refreshes on its own. If that token gets stolen, there’s a short window before it stops working.
If you choose the API key route, create the key from the Support Portal:
- Click your profile picture.
- Open Profile settings.
- Find your Freshservice API key below the Delegate Approvals section.
Then insert the key into the configuration file for whichever AI agent you're connecting.
Connecting Claude, Cursor, and Copilot Studio
Connecting an AI agent through Freshservice MCP Gateway is straightforward. Let’s take Claude, Cursor, and Copilot Studio.
Connecting Claude
To connect Claude, you’ll need:
- The Claude Desktop app / the Claude Code app (if connecting the corresponding apps, not the browser-based Claude)
- An active Claude account
- The Freshservice MCP server URL (https://your-subdomain.freshservice.com/mcp)
Connecting Cursor
To connect Cursor, you’ll need:
- The Cursor app installed.
- An active Cursor account.
- The Freshservice MCP server URL.
Connecting Copilot Studio
To connect Copilot Studio, you’ll need:
- Access to Microsoft Copilot Studio.
- An active Microsoft Copilot Studio account.
- An existing agent created within Microsoft Copilot Studio.
- The Freshservice MCP server URL.
Once these are in place, follow the step-by-step instructions on the Freshservice Support page.
Common Mistakes When Rolling Out MCP
MCP may be relatively straightforward to connect, but that doesn’t mean it should be switched on everywhere at once. A series of bad rollout decisions can quickly snowball into unnecessary troubleshooting, wasted calls, or unreliable AI output. Here are the top three mistakes to avoid.
Connecting everything on day one
Connecting all the data in the same rollout might seem efficient, but we don’t recommend it. If something goes wrong, it’s harder to identify the cause because any of the newly connected data categories could be to blame.
Ignoring rate limits
A limit of 100 calls per minute and 5,000 per month might sound generous — until an AI agent becomes part of a workflow that fires on every ticket update.
Build automation without accounting for those limits, and you may not hit the ceiling until requests start failing. Before putting a workflow into regular use, estimate how often it will trigger and how many calls it will require.
Skipping the data cleanup step
MCP surfaces your data as it stands, errors included.
Connect first and clean up later, and you may end up troubleshooting AI mistakes instead. An incorrect response caused by a duplicate asset record, outdated ticket information, or a stale knowledge base article might take longer to trace and fix than cleaning the underlying data beforehand.
Conclusion
The Freshservice MCP Gateway gives AI tools access to a wide range of IT service desk data, so connecting everything at once can be tempting. In practice, though, usage limits and data quality make a more gradual approach necessary. A phased rollout, paired with a thorough cleanup, lets you validate each connection, work with smaller high-quality datasets before the full rollout is complete, and make better use of your monthly call cap.
If your MCP rollout also involves moving from another service desk to Freshservice, Help Desk Migration can automate the transfer while prioritizing your knowledge base and highest-quality tickets. This makes key data available to AI tools sooner, without having to wait for the entire migration to finish.
Run a Free Demo Migration and see your data arrive MCP-ready.
FAQ section
- Freddy AI Agent: answering questions and tackling requests through conversational service, collaboration tools integrations, Freddy AI Agent Studio (provides customization toolkits), etc.
- Freddy AI Copilot: writing assistance (suggests fitting replies), ticket threads summaries creation, new knowledge base articles drafts, etc.
- Freddy AI Insights: forward-thinking insights, business analytics through simple conversational pieces.
Freshservice and Salesforce approach MCP servers in 2 different ways: the first as an ITSM, and the second – as an enterprise solution.
Built as an IT service desk with the focus on service desk data, Freshservice specializes in incident management, asset tracking, and ticket processing. The platform allows AI tools to surface data and query it (including searching, filtering and extracting tickets from a database). It also runs service desk tasks and workflows from external chat interfaces directly instead of having to switch software.
Salesforce, on the other hand, targets developer operations, automated business logic (Flows/Apex), and customer interactions (handles order statuses, shipping tracking, case escalations, and returns). Overall, it is designed to suit enterprise-grade environments and development-centric needs.
