satori.
    Book a call
    Back to the blog

    Anthropic MCP: connect Claude to your systems

    |7 min read
    Joachim Sahlin, CEO & AI Advisor

    Joachim Sahlin

    CEO & AI Advisor

    Pick a time for a call
    Painterly valley of dark pine forest where several winding rivers meet in one glowing channel between small cabins, illustrating Anthropic MCP connecting many business systems to Claude

    A sales manager who wants to know how the quarter is going still exports the pipeline from the CRM, pastes it into Claude and hopes the columns survive the trip. Anthropic MCP removes that step, because the Model Context Protocol is an open standard that lets Claude read and act in your own systems directly.1 Its first release on 25 November 2024 came with ready-made servers for Google Drive, Slack, GitHub, Git, Postgres and Puppeteer.1

    The protocol's own documentation compares MCP to a USB-C port for AI applications, meaning one plug that fits every device instead of one cable per brand.2 This guide explains what that means for a European company, what GDPR asks of you and how to start small.

    What is Anthropic MCP, and why does it matter?

    Without MCP, every AI tool needs its own custom connection to every system, so ten tools and ten systems give you a hundred integrations to maintain. With MCP you build one MCP server, meaning a small program that exposes chosen data and actions from one system, and every client that speaks the protocol can use it.

    Anthropic donated MCP to the Agentic AI Foundation, a new fund under the Linux Foundation, on 9 December 2025, alongside Block's goose and OpenAI's AGENTS.md.3 The same announcement counted more than 10,000 published MCP servers and named Claude, ChatGPT, Microsoft Copilot, Gemini, Cursor and VS Code among the platforms that support the protocol.3 A server you build for your CRM today therefore works with the next AI tool your company buys, whoever makes it.

    What can you do with Anthropic MCP in a normal week?

    Picture a project manager asking Claude which tickets in the tracker are overdue, where an MCP connection lets Claude fetch the open tickets, read the deadlines and answer in the chat. Nobody exports a spreadsheet, and the answer arrives in about ten seconds instead of the fifteen minutes an export and a pivot table take.

    The Microsoft 365 connector shows the same idea on tools most European offices already run. It lets Claude search SharePoint, OneDrive, Outlook and Teams in a work account, and it runs on every Claude plan.4 The account has to belong to a Microsoft Entra tenant, meaning a company directory rather than a personal mailbox. Personal accounts such as an @outlook.com address cannot connect, which catches out a surprising number of small firms.

    Four places European teams start

    Internal questions: Claude searches your handbook and policies and answers with the paragraph it used, so the HR lead stops answering the same holiday question eleven times a month.

    Finance questions: A question such as "show the ten customers who paid late last quarter" goes straight to the accounting database through a read-only server. A controller still checks the list before anyone sends a reminder.

    Project status: One question pulls ticket status from the tracker and the latest notes from the shared drive, which replaces the Monday round of messages asking where things stand.

    Actions in systems: With write access, Claude can create a ticket, update a contact or post a message. Keep write access for last, because a wrong update is harder to undo than a wrong answer. The same rule runs through our guide to AI agents with Claude.

    How do you keep MCP safe under GDPR?

    You decide exactly which tables, files and actions each server exposes, which is where GDPR's principle of data minimisation becomes a setting instead of a policy document.5 The data Claude reads through the server is still sent to Anthropic for processing, so the choice of what to expose matters more than where the server runs.

    Anthropic's commercial plans do not use your inputs and outputs for training by default, and its data processing addendum with standard contractual clauses is part of the commercial terms.67 Data from Claude Team and Enterprise is stored in the US, which belongs in your record of processing next to every other US supplier.8

    • Expose only the tables and fields the use case needs, and leave personal data out of the first pilot.
    • Grant read access before write access, and add write access one action at a time.
    • Log every call to sensitive resources, so that you can show who asked what and when.
    • Run a data protection impact assessment under Article 35 before you connect personal data at scale.5

    The protocol's own security guidance adds one rule worth repeating to any developer, which is to request the smallest set of permissions first and ask for more only when an action needs it.9 Only an Owner or Primary Owner can switch connectors on for a Team or Enterprise workspace, and each person then signs in individually, so the admin keeps the final switch.10

    How do you get started with MCP?

    Three levels cover almost every company, and each one should prove its worth before you move to the next.

    Level 1, test it yourself: Switch on one ready-made connector, such as Google Drive or Microsoft 365, and ask it the question your team asks every week. The test takes an afternoon and costs nothing beyond the Claude seat, which starts at $20 per user per month on Team billed annually.11

    Level 2, run a pilot: Pick one system and one use case, connect it through an existing or a small custom server, and count the minutes saved for thirty days. A pilot that saves less than two hours a week across the team is not worth the next step.

    Level 3, build a controlled layer: Put your most used systems behind servers with access control, logging and an owner for each one. We build this through satori-mcp at a fixed price agreed after a scoping call. satori-claude runs the Claude licences around it for €14 per user per month plus the licence, after a one-time €490 setup. All satori. prices are fixed and exclude VAT.

    Is MCP worth it for your team?

    MCP pays off when your team still copies data between systems and Claude by hand every day, and it does little for a team that only asks Claude to rewrite emails. The honest test is the copy count, since three or more copy and paste runs a day per person means a connector will save time from the first week.

    The short version

    Anthropic MCP is an open standard, now run by a Linux Foundation fund, that lets Claude read and act in your own systems. Start with one ready-made connector, keep write access for last and log every call.

    Count the times you copied text from another system into Claude yesterday, and write the name of each system next to the count. In ten minutes you have a list, and the system at the top is where your first connector belongs.

    Sources

    Footnotes

    1. Anthropic (2024). Introducing the Model Context Protocol. https://www.anthropic.com/news/model-context-protocol ↩ ↩2

    2. Model Context Protocol (2026). What is the Model Context Protocol? https://modelcontextprotocol.io/introduction ↩

    3. Linux Foundation (2025). Linux Foundation Announces the Formation of the Agentic AI Foundation. https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation ↩ ↩2

    4. Anthropic (2026). Set up the Microsoft 365 connector. Claude Help Center. https://support.claude.com/en/articles/12542951-set-up-the-microsoft-365-connector ↩

    5. European Union (2016). Regulation (EU) 2016/679, General Data Protection Regulation, Articles 5 and 35. https://eur-lex.europa.eu/eli/reg/2016/679/oj ↩ ↩2

    6. Anthropic (2026). Is my data used for model training? Anthropic Privacy Center. https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training ↩

    7. Anthropic (2026). How do I view and sign your Data Processing Addendum? Anthropic Privacy Center. https://privacy.claude.com/en/articles/7996862-how-do-i-view-and-sign-your-data-processing-addendum-dpa ↩

    8. Anthropic (2026). Where are your servers located? Anthropic Privacy Center. https://privacy.claude.com/en/articles/7996890-where-are-your-servers-located-do-you-host-your-models-on-eu-servers ↩

    9. Model Context Protocol (2026). Security Best Practices. https://modelcontextprotocol.io/docs/tutorials/security/security_best_practices ↩

    10. Anthropic (2026). Use connectors to extend Claude's capabilities. Claude Help Center. https://support.claude.com/en/articles/11176164-use-connectors-to-extend-claude-s-capabilities ↩

    11. Anthropic (2026). Plans and Pricing. https://claude.com/pricing ↩

    Back to all articles
    Cookies

    We use analytics cookies to see which pages people read, and marketing cookies stay off unless you tick them under Details.

    Cookie policy