Back to Blog

What Is the Model Context Protocol (MCP)? Examples, Costs and How to Start

Henri Jung, Co-founder at Superkind
Henri Jung

Co-founder at Superkind

A dark minimal network hub with multiple ports, the orange Superkind mascot sitting on top

The Model Context Protocol (MCP) is the question that lands on the CTO’s desk the moment an AI pilot moves from a chat window to real work in SAP, DATEV, SharePoint and the CRM. The question is simple: do we build a connector for every tool, or is there a standard way to plug an AI agent into our systems? MCP is that standard. Here is the plain answer, as of October 2026: what MCP is, where it fits for a business, what it costs and how to start without a protocol team.

Key takeaways

MCP is an open standard that lets AI agents talk to tools and data through one shared interface, published by Anthropic on 25 November 2024 and now adopted by OpenAI, Google, Microsoft and AWS12.

The point for a business: fewer bespoke connectors to maintain, portability across AI models and a cleaner security story, because the integration layer is a protocol, not a one-off script.

Most teams do not build MCP servers themselves. Platforms with ready-made integrations (Superkind: 1,500+ integrations such as DATEV, SAP Business One, HubSpot, SharePoint and lexoffice) take the integration work off your IT team.

Get a ready-made AI team that already speaks to your systems - Superkind brings an AI team (Mail Agent, Report Agent, Procurement Agent) that runs automatically. Start for free with unlimited users, or build your own.

Keep control and keep the data in Europe: Superkind is hosted and processed in Europe, GDPR compliant, aligned with the EU AI Act, and never trains on your data. You decide which steps wait for your OK.

Your AI team that already speaks to your systems

Re: Order 48217 - delivery?
Lena Weber, Weber & Co.
Hi, when will order 48217 ship? We planned the installation for next week.
Mail AgentAI employee
New customer email in sales inbox
Draft ready. Checked HubSpot (paid), SharePoint (goods picked) and DATEV (invoice 2026-0417 posted). Shipping on 30 Oct with DHL (tracking in the file). Reply drafted in your company tone, in German, matching the sales style guide.
Order statusPaid, picked, shipping 30 Oct
Systems checkedHubSpot, SharePoint, DATEV
ToneMatches sales style guide
Send reply to customer"Guten Morgen Frau Weber, Ihre Bestellung 48217 geht am 30. Oktober mit DHL raus ..."ApproveReject
Example: Mail Agent at work in Outlook

What Is the Model Context Protocol (MCP)?

The Model Context Protocol (MCP) is an open standard that lets AI applications and agents connect to external tools, data sources and systems through a single, shared interface. Instead of writing a new connector for every tool an AI should use, developers build to the protocol once, and any MCP-compatible AI can use the same integration. Anthropic open-sourced MCP on 25 November 2024 and published the specification, the SDKs for several languages and a reference library of MCP servers1. One example of what sits on top of this problem for a business team is Superkind: an AI team that already speaks to 1,500+ systems, hosted in Europe, so most companies never need to run their own MCP servers.

The one-line test

If the AI could not reach your systems at all before, MCP is how it reaches them cleanly. If the AI already reaches your systems through a product like Superkind, MCP is the plumbing under the floor - useful to know, but not something you own.

How it works with Superkind: your AI team for MCP-era integrations

Most companies do not want to run a protocol team. They want AI that opens the right system, picks up the right data and does the task. Superkind is the AI employee for your team: you write to Superkind in Outlook or any email, Microsoft Teams, Slack or the Superkind web app like you would to a colleague, and it does the task in your programs through 1,500+ integrations such as DATEV, SAP Business One, HubSpot, SharePoint, Excel, Salesforce, lexoffice and Google Drive. Models from OpenAI, Anthropic, Mistral and Google are available, all processed in Europe. Superkind is hosted and processed in Europe, GDPR compliant, aligned with the EU AI Act, and never trains on your data.

You do not need to build MCP servers to put AI to work. Superkind brings ready-made AI employees that already talk to 1,500+ systems, hosted in Europe. They run on their own in the background, and you decide which steps wait for your OK. Use them right away, adapt them, or build your own.

Mail Agent

AI employee for Outlook and Gmail

Customer questions arrive in the shared inbox all day, and every answer needs the order status, the invoice and the ticket history. The Mail Agent reads each new mail, looks up HubSpot, SharePoint and DATEV through Superkind integrations, drafts the reply in your company tone and hands it to you to approve. No one builds a bespoke connector for this.

Works inOutlookGmailHubSpotSharePoint+ 1,700 more
Re: Order 48217 - delivery?
Lena Weber, Weber & Co.
Hi, when will order 48217 ship? We planned the installation for next week.
Mail AgentAI employee
New customer email in sales inbox
Draft ready. Checked HubSpot (paid), SharePoint (goods picked) and DATEV (invoice 2026-0417 posted). Shipping on 30 Oct with DHL (tracking in the file). Reply drafted in your company tone, in German, matching the sales style guide.
Order statusPaid, picked, shipping 30 Oct
Systems checkedHubSpot, SharePoint, DATEV
ToneMatches sales style guide
Send reply to customer"Guten Morgen Frau Weber, Ihre Bestellung 48217 geht am 30. Oktober mit DHL raus ..."ApproveReject
Example: Mail Agent at work in Outlook

Report Agent

AI employee for Teams and Excel

Monday mornings used to start with Daniel pulling numbers from DATEV, SAP Business One, HubSpot and the webshop and pasting them into one deck. The Report Agent runs the same route every Monday at 07:00 across those systems, writes the short commentary and posts it to the leadership channel in Teams. No one has to maintain custom tool wrappers.

Works inMicrosoft TeamsExcelDATEVHubSpot+ 1,700 more
Weekly reportLeadership
Report AgentAI employee
Routine · every Monday 07:00
Monday report ready. Revenue week 44 EUR 318,420 (+6% w/w), 47 new orders, 3 overdue invoices flagged, pipeline up by EUR 112,000. Numbers pulled from DATEV, SAP Business One, HubSpot and the webshop, commentary written in your voice.
RevenueEUR 318,420 (+6%)
Overdue3 invoices, EUR 24,110
Pipeline+ EUR 112,000
Runs automatically every Monday at 07:00
Example: Report Agent at work in Microsoft Teams

Procurement Agent

AI employee for Slack and SAP

Quotes come in by email from different suppliers, each in a different format, and the buyer wants a clean comparison before the Monday call. The Procurement Agent collects the quotes from Outlook, pulls the matching article from SAP Business One, builds the side-by-side table and posts it to the purchasing channel in Slack. Over the sign-off threshold, management approves before the order goes out.

Works inSlackOutlook+ 1,700 more
##purchasing
Procurement AgentAI employee
New quote email in purchasing inbox
Three quotes for article FR-220 compared: Supplier A EUR 48.90/unit, Supplier B EUR 51.20/unit, Supplier C EUR 47.70/unit. Supplier C below last price (-3%), delivery in 7 days. Over EUR 15,000 total, so it waits for your OK.
Best priceSupplier C, EUR 47.70/unit
Order valueEUR 23,850
Delivery7 days
Approve order with Supplier COver EUR 15,000 - needs management OK before the order is sent in SAP Business One.ApproveReject
Example: Procurement Agent at work in Slack

Every employee is ready to use right away. Or build your own on the platform.

Safe by default

Whether the Mail Agent reads a customer email in Outlook, the Report Agent pulls numbers from DATEV every Monday or the Procurement Agent compares quotes in SAP Business One, the data stays in Europe, is never used for training, and sensitive steps wait for your approval.

Send Superkind your first task

Start for free in Outlook, Teams or Slack. 1,500+ integrations ready, hosted in Europe, approval where you want it.

A neatly coiled bundle of dark patch cables with a single coral-orange tip on one connector

How MCP Works

Under the hood, MCP uses a client-server design on top of JSON-RPC 2.0 messages. The AI application acts as a client; each external system is exposed through an MCP server. The 2026-07-28 specification revision made the protocol stateless, so each request carries its own metadata and remote servers behave like any other HTTP workload35. In practice, four parts make it up:

  1. MCP host - the AI application the user talks to, for example an AI employee, an IDE assistant or a desktop agent.
  2. MCP client - the part of the host that speaks the protocol and manages one connection per server.
  3. MCP server - the piece of software that exposes a system (a CRM, an ERP, a drive, a database) as a set of tools, resources and prompts the AI can use.
  4. Transport - how client and server talk: locally over standard input/output for desktop tools, or over HTTP for remote servers.

The server advertises three things: tools (functions the AI can call, like “create ticket” or “post invoice”), resources (data the AI can read, like “this customer” or “this document”) and prompts (reusable templates the AI can request). The protocol handles discovery, authorisation and streaming. For a business reader, the key point is that these pieces separate the AI model from the integration: the model does not need to know how to talk to DATEV; it just needs an MCP server that does.

Why it matters now

The protocol crossed from an Anthropic project to a shared standard in 2026, with OpenAI, Google, Microsoft and AWS shipping native support, governance moved to the Linux Foundation’s Agentic AI Foundation and public server registries listing thousands of implementations46. That means the integration you build or buy today is more likely to still work in two years.

MCP in Comparison

The word “integration” covers at least four very different approaches. The table below puts MCP next to the older options so you can see where each one stops. As of October 2026.

AspectMCP serverCustom API connectoriPaaS / Zapier-styleRPA / screen bots
What it doesExposes a system as tools and data for AIOne-off code glue between two systemsTrigger-action flows between appsReplays screen clicks and keystrokes
Built for AI agentsYes, from day oneNo, has to be re-wrappedPartial, newer platforms adding itNo, pre-dates the agent era
Portable across modelsYes - one server, many clientsNo - tied to the app and modelPlatform-specificNot applicable
MaintenanceYou (or a hosted provider)You, per connectorPlatform (billed per flow or seat)You, breaks when screens change
Who runs it in your companyYour IT or your AI platformYour developersAn ops/automation ownerAn RPA team
Hosting / Data protectionDepends on who runs the serverDepends on the stackVendor hosting, often USOn your machines

The honest caveat: MCP is a protocol, not a product. It gives you a cleaner way to wire AI into tools, but someone still has to run the servers, patch them and watch who sees the data. That is why most Mittelstand teams use MCP indirectly, through a platform like Superkind that keeps the integration layer managed, hosted in Europe and audited. See also MCP Connectors Explained, Zapier alternatives and agentic vs generative AI.

Examples: MCP in Daily Work

Six situations an IT lead or CTO runs into, and what MCP changes about each.

1. Superkind replies to customer email with the full file

A customer asks about an order in Outlook. The Mail Agent reads the mail, looks up HubSpot for the customer, SharePoint for the goods status and DATEV for the invoice, and drafts the reply. Underneath, it uses integrations across all three systems - the MCP-era benefit is that the same AI can pick up a new system next month without a code project. See email inbox triage for how this runs day to day.

2. A reporting agent that reads DATEV, SAP and the webshop

Instead of three Monday-morning data pulls, the Report Agent runs a routine every Monday at 07:00 across DATEV, SAP Business One, HubSpot and the webshop. For an in-house team, that would mean a connector per system; with Superkind, the integrations ship ready to use and only the routine itself needs to be set up.

3. A procurement agent that compares supplier quotes

Quotes arrive by email in different formats. The Procurement Agent collects them, looks up matching articles in SAP Business One and posts a side-by-side table to Slack. Over your sign-off threshold, management approves before the order goes out.

4. An IDE assistant that uses an in-house MCP server

A platform team builds its own MCP server for an internal documentation system and plugs it into their coding assistant. This is the classic developer case for MCP: a shared server replaces every engineer writing their own copy of the connector. The protocol here is useful because the users are already developers.

5. A ChatGPT or Claude Code setup with a few public MCP servers

A small team wires up public MCP servers (for GitHub, a database, a filesystem) into their coding assistants. This works well for development tasks. For business workflows that need DATEV, SAP, HubSpot and your email, with approvals and audit, a managed platform is the easier starting point.

6. An enterprise with its own MCP gateway

A larger IT organisation runs an MCP gateway so every internal AI agent uses the same authenticated connection to its ERP. This is useful when the company already has the platform team to run it. For teams without that in-house capacity, buying the integration layer tends to beat building it, especially under the EU AI Act and GDPR.

What MCP Costs (and What You Need)

The protocol itself is free and open source. The cost sits in two places: building and running servers, or using a platform that already has the integrations.

Option A: Build and run your own MCP servers

  • Compute is cheap - a small VPS to host an MCP server costs under EUR 10 a month, and public cloud equivalents sit in the same low range for small workloads.
  • Development is not - public vendor estimates for a single enterprise-grade server (authentication, logging, governance) run from thousands to tens of thousands of euros per integration, before upkeep. One analysis estimates a 20-server fleet at several thousand euros a month in operations alone.
  • Security sits with you - public CVEs on individual MCP servers from 2025 and early 2026 mean patching, token management and vulnerability monitoring become ongoing work.
  • You still need the agent - the server exposes the system; something else (an agent, a client, a product) uses it.

Option B: Use a platform that ships the integrations

  • Superkind pricing (as of October 2026, prices excl. VAT):
    • Free - EUR 0 with a one-off up to EUR 150 in credits to start. No credit card, no sales call, every feature and connection, unlimited users, hosted in the EU.
    • Business - from EUR 50 per month for 25,000 credits (EUR 100 for 50,000 credits is the popular step). Unlimited users, no price per seat, unused credits carry over one month, custom apps included, cancel monthly.
    • Enterprise - custom price for your volume, dedicated support with SLA, SSO/SCIM/audit log, DPA and help with the security review, pay by invoice.
  • What credits buy - a question 10-50 credits, a task 50-500 credits, a run of your own agent 1,000-5,000 credits. “EUR 100 covers up to 1,000 tasks or 5,000 questions.”
  • What is included - 1,500+ integrations such as DATEV, SAP Business One, HubSpot, SharePoint, Excel, Salesforce, lexoffice and Google Drive, plus the AI employees that use them. Models from OpenAI, Anthropic, Mistral and Google, processed in Europe.

Data protection and EU hosting

For a protocol that moves data between an AI model and your real systems, where the servers run is a first-order question. With Superkind, every task - a customer email, an invoice, a quote comparison, a weekly report - is processed in Europe from the first step, never used for training, and sensitive actions wait for your approval. That is the piece a bare MCP spec does not solve for you.

How to Start

You do not need a protocol strategy to put AI to work this week. Start with one task that crosses two systems.

Your first week

  • Give Superkind one real task - sign up for free at app.superkind.ai with up to EUR 150 in credits, forward one customer email in Outlook and let Superkind look up the order status in HubSpot or SAP Business One through the 1,500+ integrations.
  • Watch where the approval lands - decide which steps need your OK (sending a reply, posting an invoice, placing an order) and which run on their own.
  • Add a routine - turn one Monday task into a schedule (a weekly report, an overdue-invoice sweep), so Superkind keeps running without a new briefing.
  • Only build your own MCP server if you have a system Superkind does not already cover and a developer team to maintain it; otherwise the ready integrations win on time and total cost.
  • Document the setup - add the AI tools you use to your record of processing and brief staff on how to use them responsibly (Art. 4 EU AI Act7).

From the First Task to Your Own App

The real value of a protocol like MCP is that your setup does not reach a ceiling. With Superkind, the first task is where it starts, not where it ends. Four stages, on the procurement example from above:

  • Start - Ready in minutes. Sign up for free, forward one supplier quote in Outlook and watch Superkind compare it against the article in SAP Business One.
  • Relief - Tasks become routines. What you explain once keeps running: every quote comes in, lands in the comparison table and waits for your approval over the sign-off threshold.
  • Knowledge - Learns from day one. Every task makes Superkind smarter. Suppliers, prices, lead times and the sign-off rules end up in your Company Brain, and that knowledge is never lost.
  • Growth - When it gets more complex, you don’t start over. Build custom solutions, apps and integrations on top of what Superkind already knows - for example your own procurement-approval app on the platform, your Super App. Built yourself or with the Superkind team. It all belongs to you.

Conclusion

The Model Context Protocol is the right answer to a real problem: every AI tool used to need its own custom connector, and that cost did not scale. MCP replaces that with a shared standard, backed by Anthropic, OpenAI, Google, Microsoft and AWS and now governed through the Linux Foundation. For a business, the honest takeaway is this: you benefit from MCP, but you probably do not need to build on it. Superkind runs the integration layer for you, with 1,500+ connections to DATEV, SAP Business One, HubSpot, SharePoint and more, hosted in Europe, with models from OpenAI, Anthropic, Mistral and Google. Next step: send Superkind one real task from your inbox today - and leave the protocol work to the people who like protocols.

Frequently Asked Questions

MCP is an open standard that lets AI agents and models talk to tools, files and systems through one shared interface. Instead of writing a custom connector for every tool, developers build to the protocol once, and any MCP-compatible AI can use the integration. Anthropic published MCP as an open standard on 25 November 2024, and OpenAI, Google, Microsoft and AWS have since adopted it. For a business, the point is lower integration cost and portability across models.

MCP was created at Anthropic by David Soria Parra and Justin Spahr-Summers and open-sourced on 25 November 2024. The specification, SDKs and server reference implementations are maintained by the Model Context Protocol project on GitHub. Governance moved to a Linux Foundation directed fund in December 2025, so the protocol is no longer controlled by a single company.

No. MCP matters for teams that build their own agents and want to plug them into many tools without writing custom connectors each time. If you want AI employees that already work in Outlook, Teams, Slack and 1,500+ business systems like DATEV, SAP Business One and HubSpot, you can use a platform like Superkind without touching MCP yourself. Superkind handles the integration layer so you focus on the task.

Superkind is the business-ready layer on top of the integration problem MCP solves. Your team writes to Superkind in Outlook, Teams, Slack or the web app, and it does the task in your programs through 1,500+ integrations such as DATEV, SAP Business One, HubSpot, SharePoint, Excel, lexoffice and Salesforce. Models from OpenAI, Anthropic, Mistral and Google are available through Superkind, processed in Europe. You get the portability without maintaining your own MCP servers.

The protocol itself is free and open source. The cost sits in two places. First, if you build and host your own MCP servers: raw compute is cheap (a VPS from under EUR 10 a month), but staff time for authentication, patching, certificates and vulnerability management adds up fast, with vendor estimates putting a 20-server fleet at thousands of euros a month in operations. Second, if you use a platform that ships the integrations: Superkind starts free with a one-off up to EUR 150 in credits, Business from EUR 50 per month for the whole company with unlimited users (as of October 2026).

MCP is a protocol, not a hosting choice, so safety depends on who runs the servers and where. Public reporting lists CVEs in individual MCP servers from 2025 and early 2026, so the right question for IT is: where does the server run, who patches it and who sees the data. Superkind runs the integration layer for you, is hosted and processed in Europe, is GDPR compliant, aligns with the EU AI Act, and does not train models on your data. The DPA comes through the Trust Center.

The opposite. One of MCP's stated goals is to replace vendor-specific connectors with a shared standard, so your integrations keep working if you switch models. With Superkind, you already use models from OpenAI, Anthropic, Mistral and Google side by side, chosen for the task, and switching models does not break the integrations.

Henri Jung, Co-founder at Superkind
Henri Jung

Co-founder of Superkind, the AI employee teams write to in Outlook, Teams and Slack. Henri helps companies, from small teams to the Mittelstand, hand their routine work to AI - without new tools, consultants or an in-house AI team.

Send Superkind your first task.

Start for free in Outlook, Teams or Slack with 1,500+ integrations ready. Or see in 30 minutes how it runs in your systems.