Skip to content

MCP (Model Context Protocol)

| Article: 4 min read | AI in e-commerce

MCP (Model Context Protocol) is an open standard that lets AI applications (Claude, ChatGPT and others) connect securely to external systems: files, databases, CRMs, spreadsheets, services. The official documentation compares it to USB-C for AI: one type of connector instead of a separate cable for every device.

Explained simply

A language model on its own knows only what is in the conversation. For it to look up an order in your accounting, find a customer in the CRM or update a product card, it has to be connected to those systems. It used to be that a separate integration was written for each pair of "AI application + system". MCP sets a common language: a system describes once what it can do ("find an order", "create a lead"), and any AI application that supports MCP can use it.

The MCP documentation says that through it AI applications can connect to data sources (local files, databases), tools (search, calculators) and workflows, and among the examples: an agent gets access to Google Calendar and Notion, enterprise chatbots work with several databases (modelcontextprotocol.io).

Where it came from and how widespread it is

Anthropic introduced MCP on 25 November 2024 as an open standard for connecting AI assistants to systems where data is stored (Anthropic, 2024). That is the history of the term. On 9 December 2025 Anthropic donated MCP to the Agentic AI Foundation under the Linux Foundation, co-founded by Anthropic, Block and OpenAI, with support from Google, Microsoft, AWS, Cloudflare and Bloomberg. The announcement cites more than 97 million SDK downloads a month and 10,000 active servers, and support in ChatGPT, Claude, Cursor, Gemini, Microsoft Copilot and Visual Studio Code (MCP blog, 2025). These are the project's own figures.

What it looks like for a store

Examples of what can be connected (technically possible; check the readiness of specific services separately):

  • CRM: ask "show me leads from the week with no reply" or "create a task for the manager". See CRM for an online store.
  • Accounting and warehouse: "which products run out within 7 days".
  • Catalog: find products without photos or with empty specifications, prepare edits (PIM, AI for catalog content).
  • Site or store admin: change a text, update a price on request, if the system allows it.
  • Analytics: a sales summary from your exports.

MCP is also used in a new direction, where a store exposes its capabilities to shopping agents. For example, Google describes how businesses can connect to the Universal Commerce Protocol via API, A2A or MCP (Google Developers Blog, 2026). See agentic commerce.

About security

By connecting an AI to a system, you give it a key to data and actions. The official MCP security section describes typical risks:

  • Overly broad permissions. A token with access to everything, if an attacker gets hold of it, lets them do everything. Recommendation: start with minimal permissions (read) and expand gradually (MCP, Security Best Practices).
  • Local servers from an unknown source. A program that runs on your computer with your rights can execute arbitrary commands. The client should show the exact command and ask for consent; the server is best run in an isolated environment (same source).
  • Authorization errors between services (the "confused deputy" problem): an intermediary server without proper consent checks can hand access to an attacker (same source). This is a question for the server's developers, but it matters for you to take servers from trusted authors.

Practical rules for a store owner:

  1. Connect only servers from the system's vendor or a verified author.
  2. Give read permissions first; add write and delete separately and for specific actions.
  3. For payments, refunds and bulk changes, require human confirmation.
  4. Keep an action log and keep backups.
  5. Don't paste passwords and keys into the chat: use the system's own authorization mechanisms.

Common mistakes

  • Giving the AI administrator access "so that it works".
  • Installing random MCP servers from the internet.
  • Allowing changes in the live database without a backup.
  • Thinking MCP makes the AI smarter. It only provides access; the quality of decisions depends on the model and your oversight.

FAQ

What is MCP in simple terms? A single "connector" for plugging an AI into your systems: CRMs, databases, spreadsheets, services.

How does MCP differ from an API? An API is the way one program calls another. MCP standardizes how an AI application finds out which tools a system has and calls them. Often, inside, the MCP server itself uses the system's ordinary API.

Is it safe to connect an AI to a CRM through MCP? As safe as the permissions are limited and the servers verified. The official MCP documentation advises minimal permissions, user consent for dangerous actions and isolation of local servers.

Who created MCP and who supports it? Anthropic, in November 2024. Since December 2025 the project has been under the Linux Foundation; it is supported by, among others, Claude, ChatGPT, Gemini, Microsoft Copilot and Visual Studio Code.

Does a small store need MCP? Not necessarily. It is useful when you want an AI to work with your live data rather than copies in a chat. To do that, check whether your systems have an MCP server.

Sources

Want to check this on your store?

AI automation

An AI assistant checks the store every day, suggests fixes and makes them after your «Yes».

About the service →

Share

Related

  • Agentic commerce

    | Article: 5 min read

    Agentic commerce is shopping in which an AI agent acts on a person's behalf: it searches for and compares products, and within the permission it has been given it can also place the order and pay. For a store, this means part of the "visitors" will be…

  • Technical debt

    | Article: 6 min read

    Technical debt is the future cost of fixing and reworking that arises when software was built fast and without room for change. Like financial debt, it lets you move faster today, but with interest: every next change costs more and takes longer.

  • Vibe coding

    | Article: 4 min read

    Vibe coding is building programs where a person describes the task in ordinary language, AI writes the code, and the person accepts the result almost without checking it and goes by whether it "looks like it works", adding new wishes in the chat. For quickly…

  • How an online store team uses AI every day

    | Article: 2 min read

    AI in a store team's daily work is a set of small, repeating tasks in which a language model prepares a draft and a person checks it and decides. The value comes not from separate "smart" features but from simple habits assigned to specific people.