Customer service · MCP connections through the gateway

An AI package for the whole support team

Your ticketing system, your customer history and your help articles are connected once in the gateway. After that the team asks in its own words, and nothing reaches a customer without a human saying yes.

How the gateway works
Your toolsThe Walma gatewayThe modelsYour own agents
121MCP servers we have surveyed
68official, straight from the vendor
16ready to connect for support
1login for the team
Why Walma

Why go through Walma instead of straight to the model?

One account is fine for one person. The difference shows up when a whole team uses AI, and when somebody asks what it cost or where the data went.

Everyone on their own

Everyone signs up separately. Nobody knows what it costs, who has access, or what has been shared.

With Walma

One way in to every model. One login, one invoice, and you can see what is actually happening.

01 · Freedom

Do not marry one model

Claude today, something else next year. With Walma you use all of them and switch when it suits you, without redoing the work.

  • The same connections whichever model you run
  • Switch model without rebuilding anything
  • Run two models on the same task and compare
02 · Ready

The connections are already built

We have done the groundwork: ready-made connections to the tools you already use, with proven skills and instructions, so the team actually gets going instead of just trying it out.

  • Ready connections to the tools you already use
  • Skills and instructions other teams already run in earnest
  • Nothing for the team to install
03 · Control

You see exactly what the AI does

If you want visibility, you get it: everything the agent does is visible, you get feedback on how the work could be better, and you can put a stop to things it should not do.

  • Every call logged, with person, tool and timestamp
  • Feedback on how the team could use it better
  • Guardrails for what the agent must not touch
How the AI Hub works

Three steps, and then it is running

Walma AI Hub is a layer between your tools and the models. Here is what that means in practice.

01Step 1

Connections are approved once, centrally

IT approves an MCP server once. After that the tool appears for the people who should have it, with the right permissions. Nobody on the team touches an API key.

  • One allowlist instead of config files on every laptop
  • Read or write permissions per role, set in the gateway
  • Revoke access in one place when someone leaves
  • Gorgiasbeta
  • Intercombeta
  • Freshdeskofficial
  • Help Scoutofficial
02Step 2

Every model behind the same key

Claude, GPT, Gemini and the open models all reach the same connections. You pick the model per task instead of locking yourself in, and if you switch next year the tools come along.

  • One login for the user, whichever model runs
  • Budgets and spend caps per team, not per person
  • One log for model calls and tool calls alike
03Step 3

Your own agents on top, built by us

When the off-the-shelf tools run out, we build. RAG over your own material, bespoke agent flows, and narrow MCP servers for systems the vendor has not covered, all running inside the same gateway and the same log.

  • RAG over your documents, contracts and product data
  • Agent flows that follow your process, not a template
  • Custom MCP servers for systems the vendor has not covered
Your material
Documents and contracts
Product and customer data
Internal systems
The agent we build
Walma agentRAG · tools · your process
Answers in Claude, Teams or Slack, where people already work
Connections

What support gets to connect

This category is moving fast. Freshdesk and Help Scout ship their own servers, Zendesk chose to build a client instead, and for the rest we build the connection.

Ready to connect

The vendor's own server. Switched on and running the same day.

  • Gorgiasbeta
  • Intercombeta
  • Freshdeskofficial
  • Help Scoutofficial

Needs review first

Community-built. We review the tool descriptions and pin the version before approving.

  • Zendeskcommunity
  • Zoho Deskcommunity
  • LiveAgentcommunity

In progress: we build the connection

These we build ourselves, as narrow read-only servers in your tenant. Tell us which you need first.

  • Frontin progress
  • Kustomerin progress
  • Dixain progress

Base tools, whatever the department

Included in every rollout.

  • Google Workspaceofficial
  • Microsoft 365beta
  • Slackofficial
  • Notionofficial
  • Atlassianofficial
  • Linearofficial
  • Asanaofficial
  • Boxofficial
  • Dropboxofficial
  • Zapierofficial
  • Makeofficial
  • n8nofficial
In practice

What you actually ask it

Questions support already asks, minus building a report first.

  • → Summarise this week's tickets and what customers complain about most.
  • → Which tickets have been open longest, and who are we waiting on?
  • → What has this customer contacted us about before, and how was it resolved?
  • → Draft a reply to this ticket, in our tone and with the right help article.
  • → Which questions come up often enough that they should be a help article?
  • → How many tickets were about delivery delays this month?
How it fits together

It all runs in the EU, with a log

The gateway runs in your own Azure tenant in an EU region. Every call is logged with user, tool and arguments. That is the answer when your data protection officer asks where the data goes.

Security

Built for European requirements from day one. Details, subprocessors and policies live in our Trust Center.

Open the Trust Center
  • GDPR
  • EU data residency
  • Encrypted at rest & in transit
  • No training on your data
  • Tenant isolation
  • SSO & SCIM
  • SOC 2 Type IIin progress
  • ISO 27001in progress
  • Custom user roles
FAQ

What support managers ask first

Does the AI reply to customers directly?+

Not with us by default. Connections are set read-only, so the agent reads tickets, finds the history and drafts replies, but a human sends. If you want automatic replies on simple tickets that can be opened up, as an active decision per tool with approval and a log. A wrong answer to a customer cannot be taken back.

Can Zendesk be connected?+

Not through a server of Zendesk's own, because they chose a different route: they are building an MCP client instead, so their product consumes other servers rather than exposing one. Access the other way therefore goes through an app or a community-built server, which we review before approving.

Which ticketing systems can be connected today?+

Freshdesk has an official Freshworks server, generally available since 10 September 2026, and Help Scout has its own. Gorgias is in open beta and Intercom in beta. Zoho Desk and LiveAgent are available through community and aggregator routes. For Front, Kustomer and Dixa we build the connection ourselves.

We run Intercom, does being in the EU matter?+

Yes, and it is worth knowing before you plan. Intercom's server is read-focused and supports US-hosted workspaces only. If your workspace is in the EU it does not reach it. The gateway itself runs in your own Azure tenant in an EU region, but we cannot conjure away a vendor's limitation, and we would rather say so now than during a rollout.

Can the AI see what customers have written?+

Only what you open up, and only for the people who should see it. Permissions are set per role in the gateway. Bear in mind that ticket text is content written by someone outside your organisation, which makes it a possible route in for prompt injection. That is why we run read-only sessions for that kind of data and inspect tool results at the gateway.

Connections

See it on your own tickets

Twenty minutes with an engineer. We connect your ticketing system, read-only, and show what the team can ask the next day.