Freshworks MCP gateway explained: what it is and how it works
Stevia Putri
Katelin Teen
Last edited May 15, 2026
On May 14, 2026, Freshworks announced its MCP Gateway at the annual Refresh 2026 conference in San Mateo. The announcement landed alongside the new Freddy AI Agent Studio and positioned Freshworks squarely in the middle of a broader enterprise race - ServiceNow, Atlassian, and Salesforce all shipped competing agent platforms in the same quarter.
The core problem MCP Gateway addresses is one that any IT team or developer will recognize: your AI tools are smart, but they can't see your service data. A developer using Cursor can't query a live Freshservice ticket without leaving their IDE. A Freddy AI agent that wants to update a Linear issue has to wait for someone to wire up a custom integration. Most enterprise AI still works in silos, and the cost of connecting those silos has been bespoke code and maintenance.
MCP Gateway is Freshworks' answer to that. But the way it actually works - and who it's actually for - is more nuanced than the launch narrative suggests. This post explains the mechanics, what's available on which plan, and what to keep in mind before getting excited.
What is the Model Context Protocol?
Before getting into what Freshworks built, it helps to understand what MCP is. The Model Context Protocol is an open standard for connecting AI applications to external data and tools. Anthropic developed it; it's now maintained as an open specification with support across Claude, Cursor, VS Code, GitHub Copilot, and Microsoft Copilot Studio.
The official analogy is a USB-C port: just as USB-C lets you connect any compatible device without thinking about the cable, MCP lets any compatible AI client connect to any MCP server without custom integration code. A single MCP server can serve Claude, Cursor, and VS Code simultaneously, and a new AI tool that supports MCP can plug in without any changes on the server side.
For ITSM platforms like Freshservice, the opportunity is significant. Ticket data, asset records, and knowledge base articles become accessible to any AI that speaks MCP. Developers can query live service data inside their IDE without leaving their workflow. Freshdesk AI agents can take actions in external tools without bespoke connectors. The "build once, integrate everywhere" promise has real weight for large toolstacks.
An important detail from Freshworks' own engineering blog: they don't let AI tools write arbitrary queries against their data. Each MCP tool encapsulates pre-approved logic - hardcoded queries, API calls, filtered lookups - behind a simple interface. This prevents both SQL injection risk and AI-hallucinated queries from hitting production data.
How the Freshworks MCP Gateway works
The Freshworks MCP Gateway has two modes that address different directions of data flow.
Inbound MCP runs from external AI tools into Freshservice. A developer using Cursor or Claude can connect to their Freshservice account's MCP endpoint at https://<your-freshservice-domain>/mcp and immediately start querying tickets, assets, knowledge base articles, users, and more. No custom code. Authentication is either OAuth 2.0 (recommended) or API key. The result is that an AI assistant inside your IDE has the same context about your IT environment that a support agent looking at the Freshservice dashboard would have.
Outbound MCP runs the other direction: Freddy AI agents inside Freshservice can take actions in tools outside Freshworks. Supported at launch are Atlassian products, Notion, Linear, and ClickUp. So a Freddy AI agent resolving a service request can create a Jira issue, update a Notion project doc, or mark a Linear task complete - through a single governed gateway rather than four separate integrations.
The gateway itself adds governance that a raw MCP server doesn't. Every connection goes through role-based access controls. Tool calls are rate-limited. Authentication is centrally managed. This matters at enterprise scale, where an uncontrolled AI tool hammering a live service API is a real operational risk.
The MCP tools exposed
The current Freshservice MCP server exposes 28 tools across six resource categories. Here's the full reference:
| Resource | Tools |
|---|---|
| Tickets | fetchTickets, fetchTicket, createTicket, updateTicket, createTicketNote |
| Assets | fetchAssets, fetchAsset, createAsset, updateAsset |
| Users and agents | fetchAgents, fetchAgent, fetchRequesters, fetchRequester |
| Onboarding / offboarding | fetchOnboardingRequest, fetchOnboardingRequestForm, createOnboardingRequest, fetchOffboardingRequests, createOffboardingRequest |
| Service catalog | placeRequestServiceCatalogItem, fetchServiceCatalogItemSearch, fetchServiceCatalogItem, fetchServiceCatalogItems |
| Knowledge base | fetchSolutionCategories, fetchSolutionCategory, fetchSolutionFolders, fetchSolutionFolder, createSolutionFolder, fetchSolutionArticles, fetchSolutionArticle, createSolutionArticle, updateSolutionArticle, fetchSolutionArticleSearch |
| Workspaces | fetchWorkspaces |
The coverage is solid for day-to-day ITSM use. A developer who wants their IDE to know the status of an open ticket, the assigned agent, and the relevant knowledge base articles can get all of that through a single MCP connection.
How to connect your AI tool
The setup process is different for each client, but all use the same Freshservice MCP endpoint.
Claude (web connector): In Claude's settings, go to Settings > Connectors > Add custom connector. Enter https://<your-freshservice-domain>/mcp, click Add, then Connect, and complete the Freshservice authorization flow. After that, tools are available with configurable permissions: always allow, needs approval, blocked, or custom.
Cursor: Create or update .cursor/mcp.json with:
{
"mcpServers": {
"freshservice": {
"type": "http",
"url": "https://<your-freshservice-domain>/mcp"
}
}
}
For API key authentication, add "headers": {"Authorization": "<API-KEY>"} to the server block.
Visual Studio Code: Open the Command Palette (Ctrl+Shift+P), select MCP: Add Server, choose HTTP, enter the MCP URL, and authorize in the browser.
Microsoft Copilot Studio: In your agent's Tools tab, select Add a tool > New tool > Model Context Protocol. Enter the server URL, then use OAuth 2.0 with Dynamic discovery for authentication.
Claude Code: Run claude mcp add my-remote-server --transport stdio -- npx -y mcp-remote https://<your-freshservice-domain>/mcp, then verify with claude mcp get my-remote-server. For more detail on using Claude with MCP integrations, the Claude Code MCP integration guide covers the patterns.
Authentication for all clients requires administrator or agent-level privileges on your Freshservice account.
Plan availability and rate limits
This is the part that matters most if you're evaluating whether MCP Gateway is within reach:
Outbound MCP (Freddy AI acting on external tools like Notion, Linear, ClickUp) is available on Growth, Pro, and Enterprise plans - provided you have an active Freddy AI Agent Studio. This is the more accessible mode.
Inbound MCP (external AI tools querying Freshservice data) is Enterprise-only during the Early Access Program. If you want Claude or Cursor to read your live tickets and assets, you need to be on Enterprise.
During EAP, the following rate limits apply on Enterprise:
| Limit | Value |
|---|---|
| Tool calls per minute | 100 |
| Tool calls per month | 5,000 |
| Tool access | Full |
5,000 calls per month sounds like a lot until you realize that an active developer using Cursor against live ticket data will burn through that in a few days of heavy use. These limits will likely change as the program moves out of EAP, but they're worth noting now.
The same plan structure applies for Freshdesk, not just Freshservice: Inbound MCP on Enterprise, Outbound on Growth and above with an active Freddy AI Agent.
The business case Freshworks is making
Freshworks' own telemetry is the most interesting part of the launch narrative. According to their press release, analysis of millions of service interactions found that 47% of IT tickets are now submitted outside standard business hours. After-hours SLA rates drop by as much as 5%. Response times lag by over an hour.
They call this the "ghost shift" problem: distributed teams working after hours who are let down by AI that can't access real context about their environment. A Freddy AI agent can't help an employee set up new hardware at 11pm if it can't query the asset database to find what's available. MCP Gateway closes that gap by giving AI agents access to the full Freshservice data layer at any time.
CPO Srini Raghavan framed it this way at Refresh 2026:
"The true measure of AI's value isn't what it can do, it's what it gives back: time, focus, and the freedom for teams to stop fixing yesterday's problems and start building what's next."
- Srini Raghavan, CPO at Freshworks, Refresh 2026 launch blog
The competitive angle is direct: Freshworks positioned MCP Gateway against ServiceNow, Salesforce, and Atlassian, all of which launched agent platforms in the same quarter. The differentiator Freshworks is betting on is its unified service data layer - tickets, assets, and incidents already in one platform - which it argues gives AI immediate context that fragmented stacks can't match without data migration.
What to keep in mind
A few things that don't make it into press releases:
The community MCP server (effytech/freshservice_mcp) predates Freshworks' official gateway by about a year. It had 42 forks before Freshworks launched anything official, which tells you something about how long the demand for this has existed and also means there's a working alternative if the Enterprise plan gating is a blocker.
Inbound MCP being Enterprise-only is a meaningful constraint. Most teams evaluating Freshservice for mid-market ITSM are not on Enterprise. Outbound MCP is more accessible, but that's Freddy AI acting on external tools - which requires having Freddy AI Agent Studio already configured and working, not a trivial setup.
The 4 supported outbound tools (Atlassian, Notion, Linear, ClickUp) cover the most common developer tool stack, but if your cross-functional workflow depends on something else - Jira Service Management, GitHub Projects, Asana - you're waiting on roadmap.
Freshworks' philosophy on MCP queries (hardcoded tools, not arbitrary LLM queries) is the right security posture but also means you get a defined set of tools, not an open SQL interface to your data. That's a feature, not a bug, but it's worth understanding before assuming you can ask arbitrary questions of your service data through Claude.
Understanding MCP Gateway's architecture
The gateway sits as middleware between AI clients and Freshservice's underlying APIs. Every request passes through authentication, RBAC, and rate limiting before touching live data. Outbound requests from Freddy AI go through the same governed layer before reaching external tools.
This architecture - single endpoint, centrally managed permissions, audit trail - is what distinguishes an enterprise MCP gateway from a local stdio MCP server you'd run on a laptop. The Hacker News developer community has been building toward this pattern independently; Freshworks has productized it for their platform.
For teams already running Freshdesk automation via webhooks and rules, MCP Gateway is a different layer: webhooks fire on events, MCP responds to queries. They're complementary. A Freddy AI agent could use MCP to pull context from external tools, then fire a Freshdesk webhook to notify a downstream system. Understanding where each fits in your automation workflow saves a lot of trial and error.
If you're comparing Freshworks against Zendesk, the MCP Gateway is a differentiator worth noting - Zendesk doesn't have an equivalent native gateway at this point. Whether that matters depends on how much of your AI strategy involves plugging external tools into your ITSM data versus handling customer-facing support volume.
eesel AI for Freshdesk
If the context you're in is customer support on Freshdesk rather than IT service management on Freshservice, the MCP Gateway story is less central to what you need. The gateway is mainly a developer and ITSM tool; it's not designed to autonomously resolve customer support tickets end-to-end.
eesel AI is built specifically for that use case. You connect it to your Freshdesk account, and it learns from your past tickets, help center articles, and macros. It drafts and sends replies, handles escalation rules, and closes resolved conversations - without MCP configuration. Teams using eesel on Freshdesk have reached up to 81% autonomous resolution rates. Pricing is usage-based at $0.40 per ticket.
If you want to see how eesel compares to other AI solutions for Freshdesk, that guide walks through the options including Freddy AI Agent itself. The step-by-step guide to creating a Freshdesk AI agent is a good starting point if you want to understand what setup looks like before committing to anything.
Frequently Asked Questions
Share this article
Article by
Stevia Putri
Stevia Putri is a marketing generalist at eesel AI, where she helps turn powerful AI tools into stories that resonate. She’s driven by curiosity, clarity, and the human side of technology.