Skip to content

/articlesConcepts

What is MCP (Model Context Protocol)? A guide for enterprise teams

MCP is an open standard that connects AI apps to tools and data. How MCP servers and clients work, the risks to control, and how Botify’s MCP server works.

Updated
Published
min read
07
On this page
10
On this page10

What problem does MCP solve?

Before MCP, every AI app needed its own integration with every system. Five AI apps and ten business systems could mean fifty separate integrations, each describing its functions and passing credentials in its own way.

MCP replaces that with one protocol. You wrap a system once as an MCP server. Any app that speaks MCP can then see what the server offers and use it.

MCP does not replace APIs. An MCP server usually sits in front of an existing API and presents it as tools an AI app can find and call. What MCP standardises is the conversation between the AI app and the tool.

Who created MCP, and who runs it now?

MCP was created at Anthropic by David Soria Parra and Justin Spahr-Summers and released as open source on 25 November 2024, with the specification, SDKs and reference servers for systems such as Google Drive, Slack, GitHub and Postgres.

On 9 December 2025 Anthropic gave MCP to the Agentic AI Foundation, a fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI. The project’s maintainers keep its technical direction. By then they reported support in ChatGPT, Claude, Cursor, Gemini, Microsoft Copilot and Visual Studio Code, among others.

The specification is versioned by date. The current revision at the time of writing is 2026-07-28. It makes the protocol stateless: every request carries its own protocol version and capabilities.

How does MCP work? Hosts, clients and servers

There are three parts. The host is the AI app, such as a chat assistant or a code editor. For each server it uses, the host opens an MCP client. The server is the program that offers the tools, running on the same machine or on the internet.

Messages use JSON-RPC 2.0. Local servers usually talk over stdio. Remote servers use Streamable HTTP with an API key or a token; the MCP documentation recommends OAuth to obtain tokens.

  • Discover: the client asks the server which versions and features it supports.
  • List: the client calls tools/list and gets each tool’s name, description and input schema.
  • Choose: the host shows those tools to the model, which decides whether it needs one.
  • Call: the client sends tools/call with the arguments, and the server runs the real API call.
  • Return: the result goes back to the model for its next step or its answer.
What a server offersWhat it isExample
ToolsActions the AI app can callSearch tickets, create a draft, query a database
ResourcesData the app can readA file, a database schema, an API response
PromptsReady-made instructions for the modelA template for writing a report from the server’s data

MCP vs function calling and plugins

Function calling is a model feature: the model asks to call a function you described. MCP sits one step further out. It standardises how those functions are published, found and run across apps and networks, so one tool works with any compatible host.

Plugin systems solved the same problem for one product at a time. MCP’s gain is portability: wrap an internal system once and use it from several AI apps.

Is MCP secure? The risks to control

MCP is a protocol, not a security model. It defines how tools are described and called. Whether a call should happen is up to the host app. The specification says a person should be able to deny tool calls, and that clients must not trust a tool’s own labels, such as “read-only”, unless the server is trusted.

An MCP server is outside code with access to your systems, and its tool descriptions and results are text your model reads. Treat both with care.

  • Approval: require a person’s approval for any tool that writes, deletes or sends outside the company, and do not trust the server’s own risk labels.
  • Credentials: store server keys encrypted, limit them to one company, and never put them in prompts.
  • Least access: give the server its own account with only the permissions its tools need.
  • Prompt injection: tool descriptions and results can carry instructions aimed at the model; treat them as data.
  • Supply chain: check who maintains a server and what it can reach before you add it.
  • Log: record every call, its arguments, the approval and the result.

Botify MCP: two ways Botify uses the protocol

Botify works with MCP in both directions. It runs its own MCP server, and its assistants can use tools from MCP servers you add.

The Botify MCP server: build assistants from Claude or Cursor

Botify serves an MCP server at /mcp on your Botify address. You connect an MCP client such as Claude or Cursor with a Botify API key; OAuth 2.1 sign-in works only where it has been switched on.

Through it, the client can list your assistants and their versions, create an assistant, edit its draft, add knowledge text, send a test message, run the saved test questions and check readiness. Every call is recorded.

Publishing a version follows your workspace rules. If the rules do not allow it, nothing is published and the client gets a link to finish in the Botify console.

Adding outside MCP servers to Botify assistants

You can add an MCP server to Botify by its HTTP address, for the whole workspace or for one assistant. If the server needs a key, Botify stores it encrypted. Its tools then appear next to the built-in integrations, under the same rules.

Botify does not trust the server’s own labels. Until someone on your team reviews a tool, Botify treats it as a write action that always needs approval, so every call waits for a person. After review you set its approval mode yourself. Disabling the server stops all its tools.

Botify connects to MCP servers over HTTP, so a server that only speaks stdio has to be hosted behind an HTTP address first.

Can I connect Odoo or another ERP through MCP?

Yes. For Odoo, third-party modules on the Odoo Apps Store and open-source projects publish MCP servers. Their authors maintain them, so check them like any outside software: licence, updates and the Odoo access rights they need.

Botify connects to Odoo directly, so you do not need a third-party MCP server for Odoo. ERPNext also has a built-in integration. Other ERPs can be added through their MCP server as described above.

  • Create a separate ERP user for the integration with only the access it needs.
  • Add the MCP server and review each tool before you relax its approval.
  • Relax only read tools, such as searching records; keep create, update and delete tools on approval.
  • Test with real questions on a test database before you point the assistant at live data.

Common questions

01

Is MCP only for Claude?

No. Anthropic created it, but it is an open standard under the Linux Foundation’s Agentic AI Foundation, and many AI apps and developer tools support it. Any app that implements the client side can use any MCP server.

02

Is MCP the same as an API?

No. An API is how one system exposes its functions. MCP is a common way for AI apps to find and call such functions. An MCP server usually wraps an existing API and presents it as tools with names, descriptions and input schemas.

03

Does Botify have an MCP server?

Yes, at /mcp. Connect a client such as Claude or Cursor with a Botify API key to list, create, edit, test and publish assistants. Publishing follows your workspace rules; if they do not allow it, you finish in the console.

04

Can Botify assistants use tools from my own MCP server?

Yes. Add the server by its HTTP address. Until a tool is reviewed, every call waits for approval. After review you choose its approval mode, and disabling the server stops all its tools.

05

Do I need to write code to use MCP?

Not to use existing servers: you add them in a compatible app. You write code only to expose an in-house system, using one of the official SDKs to build your own MCP server.

Sources

  1. [1]Introducing the Model Context Protocol (Anthropic, 25 November 2024)
  2. [2]MCP joins the Agentic AI Foundation (MCP blog, 9 December 2025)
  3. [3]MCP architecture overview (modelcontextprotocol.io, revision 2026-07-28)
  4. [4]MCP specification: Tools (revision 2026-07-28)
  5. [5]MCP Server module for Odoo (Odoo Apps Store, third-party)

Where you’ll find this in Botify

Start with one assistant on WhatsApp or your website.

Book a demo and we’ll show you an assistant answering questions like your customers’. We set up the first one with you and review the results together before you add more.