In This Article

Managing a contact center usually means living in a dashboard. You click into one screen to check yesterday's missed calls, another to tweak an agent's script, a third to launch a follow-up campaign, and a fourth to see whether any of it worked. Every task is a few clicks deep, in a different place.

The Mihu MCP server collapses all of that into a conversation. Connect Mihu to an AI assistant like Claude or ChatGPT and you can simply ask: "Which agents handled calls today?", "Update the service agent to offer Saturday slots", "Call this list back tomorrow morning", "How did last week's renewals campaign do?" The assistant does it — by talking to Mihu for you.

The problem with dashboards

Dashboards are great for browsing and terrible for doing. The moment you know exactly what you want — call these forty customers, change this one line in an agent's instructions, pull this number for a report — clicking through menus is just friction. And the people who most need to make these changes, the ops managers and account leads, often aren't the developers who could otherwise script it through an API.

That's the gap the MCP server closes. It puts the full power of the platform behind plain language, so the person who knows what needs to happen can make it happen directly, without learning a UI's geography or waiting on an engineer.

What MCP is, in plain terms

MCP stands for Model Context Protocol — an open standard for connecting AI assistants to outside tools and data. The idea is simple: a service publishes an MCP server that lists the things it can do; an AI assistant connects to it as an MCP client; and from then on, the assistant can use those capabilities directly inside a conversation.

Think of it as a universal adapter between AI assistants and software. Before MCP, every assistant needed a custom integration for every tool. With MCP, any compatible assistant can pick up any MCP server and start using it. Mihu's MCP server makes the contact center one of those tools.

In one sentence

MCP lets an AI assistant operate your software for you — and the Mihu MCP server makes your entire Mihu contact center one of the things it can operate.

What the Mihu MCP server is

The Mihu MCP server is a single endpoint — https://mcp.mihu.ai/mcp — that exposes your Mihu platform as a set of tools any MCP client can use. Connect once, and your assistant can manage agents, place calls, schedule work, and read results, all scoped securely to your own account. No new app to learn, no integration to build: you bring the assistant you already use and point it at Mihu.

What it lets you do

The server exposes the core operations of the platform as conversational tools. The assistant picks the right one based on what you ask.

List & inspect agents

See every AI agent in your account and pull the full configuration of any one of them.

Update an agent

Change instructions, procedures, languages, or behavior — and the change is live immediately.

Initiate calls

Launch outbound AI calls — one number or a whole list — with the context the agent should use.

Schedule tasks & campaigns

Queue future calls and follow-ups with timing, time zones, and retry rules — a campaign in a sentence.

Manage listings

Keep the inventory, services, or catalog your agents quote from accurate and up to date.

Read analytics

Ask for call outcomes, sentiment, resolution rates, and campaign performance — answered in the chat.

A real conversation

Here's what it feels like in practice — the assistant reaches for Mihu's tools on its own and reports back.

Your AI assistant + Mihu MCP
Pull up my agents and tell me which one handles service bookings.
→ mihu.list_agents ( )   ✓ 6 agents
You have six agents. "Service Desk" handles bookings — it answers in Italian and English and books into your Google Calendar.
Make it offer Saturday morning slots too, then call back everyone who missed a service this month.
→ mihu.update_agent ( availability += Sat AM )   ✓ updated
→ mihu.schedule_tasks ( segment: overdue_service, channel: voice )   ✓ 38 calls queued
Done. Service Desk now offers Saturday mornings, and I've queued callbacks to all 38 customers overdue this month for tomorrow at 10:00. Want a summary once they run?

MCP vs. the REST API

Mihu still has a full REST API, and it isn't going anywhere — it's the right tool when you're building software. The MCP server is for a different moment: when you want to operate the platform, not program against it.

REST API — for building

  • You write code for every request and response
  • You handle auth, parsing, and error states
  • Perfect for embedding Mihu inside your own product
  • Needs a developer

MCP server — for operating

  • The assistant discovers and calls tools itself
  • You ask in plain language; it chains the steps
  • Perfect for running the contact center day to day
  • Anyone on the team can use it

The short version

Use the API to build with Mihu, and the MCP server to run Mihu. Most teams end up using both.

Who it's for

Ops & account managers

Run the day from a chat window — check performance, adjust agents, and launch callbacks without clicking through five screens or filing a ticket with engineering.

Builders & agencies

Spin up, tune, and manage agents for multiple clients conversationally — and wire Mihu into agentic workflows without writing bespoke API glue for every action.

Automation engineers

Let an autonomous agent read live call outcomes and act on them — escalate, reschedule, sync — as part of a larger workflow, with the contact center reachable as a standard MCP tool.

Connect it in Claude or ChatGPT

The Mihu MCP server works with any client that speaks the Model Context Protocol. The two most common are Claude and ChatGPT, and adding Mihu to either takes a couple of minutes. In both, you'll paste the same endpoint — https://mcp.mihu.ai/mcp — and sign in once.

Claude

Pro, Max, Team & Enterprise
  1. Open Settings → Connectors (on Team/Enterprise, an Owner adds it once under Organization settings → Connectors, then members connect).
  2. Click the + button and choose Add custom connector.
  3. Paste the URL https://mcp.mihu.ai/mcp and click Add.
  4. Authenticate in the pop-up — Mihu is now available. Use the + in any chat to enable it per conversation.

ChatGPT

Plus, Pro, Business & Enterprise
  1. Go to Settings → Apps & Connectors → Advanced settings and turn Developer mode on.
  2. Back in Connectors, choose Create / Add connector and give it a name like Mihu AI.
  3. Paste the URL https://mcp.mihu.ai/mcp and sign in with OAuth.
  4. Enable Mihu from the Developer mode tools in the composer and start asking.

In Claude, dropping in the endpoint looks like this:

Illustration of the Add-connector dialog — your live screen may differ slightly as the apps update.

Two things worth knowing

Custom connectors need a paid plan on both Claude and ChatGPT — they aren't available on free tiers. And both sign in to Mihu over OAuth, so the connection inherits exactly the permissions of the account you log in with, scoped to your tenant.

Want the full step-by-step with tool reference and auth details? Read the Mihu MCP developer guide →

Use cases in Claude, in the real world

Once Mihu is connected, managing the contact center happens in plain language. Here are three things teams actually do from a Claude conversation — each shown with a real session below.

1. Morning stand-up: "How did we do yesterday?"

Instead of opening the analytics dashboard, an ops manager just asks. Claude calls Mihu's analytics tools and answers with yesterday's call volume, resolution rate, sentiment, and anything that needs attention — in seconds.

Add your real screenshot here A Claude chat asking for yesterday's performance, with Mihu's analytics tool result rendered in the reply. /images/mcp/claude-analytics.png

2. Editing an agent without touching the dashboard

"Add Saturday morning slots to the Service Desk agent and have it mention our winter promo." Claude reads the agent, applies the change through Mihu's update tool, and confirms it's live — no menus, no redeploy.

Add your real screenshot here A Claude chat updating an agent's availability and script, showing the Mihu update-agent tool call and confirmation. /images/mcp/claude-update-agent.png

3. Launching a callback campaign from chat

"Call everyone who missed a service this month, tomorrow at 10am." Claude segments the list, schedules the outbound calls with retry rules through Mihu, and reports how many were queued — a full campaign started in one sentence.

Add your real screenshot here A Claude chat launching a callback campaign, showing Mihu's schedule-tasks tool result with the number of calls queued. /images/mcp/claude-campaign.png

The same works in ChatGPT

Every one of these runs identically in ChatGPT with Developer mode on — same endpoint, same tools, same plain-language control. Pick the assistant your team already lives in.

Security

Conversational control doesn't mean loose control. Connections are authenticated with scoped credentials and isolated to your own tenant, so an assistant only ever sees and acts on your agents, your calls, and your data — never anyone else's. You decide which assistant gets access, and you can revoke it at any time.

Your tenant, your rules

Access is tenant-scoped and key-based. The assistant operates strictly within the permissions you grant — and you can turn it off as easily as you turned it on.

Frequently asked questions

What is the Mihu MCP server?

A Model Context Protocol endpoint that exposes the Mihu platform as tools any MCP-compatible AI assistant can use. Instead of calling an API in code, you connect a client like Claude and manage your contact center in plain language — agents, calls, campaigns, and analytics.

What is MCP (Model Context Protocol)?

An open standard for connecting AI assistants to external tools and data. A service runs an MCP server, an assistant connects as a client, and the assistant can then discover and use that service's capabilities directly in conversation.

How is it different from the Mihu REST API?

The REST API is for developers writing software against Mihu. The MCP server lets an AI assistant call those same capabilities itself, in natural language, chaining steps together — no integration code to write or maintain. Use the API to build with Mihu, the MCP server to run it.

Which assistants can connect?

Any MCP-compatible client. The most common are Claude (Pro, Max, Team, Enterprise) and ChatGPT (Plus, Pro, Business, Enterprise, with Developer mode enabled), plus AI-enabled IDEs and agentic automation tools that support the Model Context Protocol.

How do I add Mihu in Claude?

Open Settings → Connectors, click +, choose Add custom connector, paste https://mcp.mihu.ai/mcp, click Add, and authenticate. On Team or Enterprise plans an Owner adds it once for the organization, then members connect individually. Full walkthrough: developers.mihu.ai/guides/mcp.

Does it work with ChatGPT?

Yes. Turn on Developer mode under Settings → Apps & Connectors → Advanced settings, add a connector pointing to https://mcp.mihu.ai/mcp, and sign in with OAuth. It's available on paid ChatGPT plans on the web.

Is it secure?

Yes. Connections use OAuth and are isolated to your tenant, so an assistant only ever accesses your own data with the permissions of the account you sign in with. You control which assistant connects and can revoke access anytime.

Connect your assistant to Mihu

Point Claude, ChatGPT, or any MCP client at mcp.mihu.ai/mcp and start running your contact center by just asking. New to it? The developer guide walks you through every step.

Get your MCP endpoint Works with Claude, ChatGPT & any MCP client · OAuth & tenant-isolated · Setup in minutes