What is the Model Context Protocol (MCP)? Complete Guide

Q: Is the Model Context Protocol free?

Yes. MCP is an open standard published under a permissive license. You pay no royalty to Anthropic to use it. Only the third-party services you connect (a Claude subscription, a cloud server, a commercial connector like Aidoo) may carry a cost.

Q: Do I need to be a developer to use MCP?

No, not to use it as an end user. If you use Claude Desktop with a preinstalled MCP server, you just ask your questions. However, building an MCP server requires development skills (Python, TypeScript, or another supported language).

You may have heard about the Model Context Protocol (MCP) without fully understanding what it is. An open standard published by Anthropic in November 2024, MCP has become in a few months the common grammar spoken by modern AI assistants to connect to your business tools. For an Odoo user (Odoo being a popular open-source ERP), this is a shift: your software is no longer a maze of screens, it becomes conversational. This guide explains plainly what MCP is, how it works, and why you should care now.

📌 TL;DR: the essentials in 30 seconds

MCP is an open standard that lets an AI assistant (Claude, ChatGPT…) use external tools (Odoo, Gmail, GitHub, etc.) without bespoke integrations per vendor.

It runs on a client-server model: your application exposes an "MCP server", the AI accesses it through an "MCP client".

MCP standardizes how tools are described, invoked and secured, think of it as a USB-C port for AI.

For Odoo, MCP turns the ERP into a conversational interface: your teams query data and trigger actions in plain English.

The ecosystem is adopting MCP rapidly through 2025–2026: Anthropic, OpenAI, IDEs, and ERP platforms.

What is the Model Context Protocol (MCP)?
Why was MCP created?
How MCP works: client-server architecture
The main MCP use cases
Security, governance and GDPR compliance
MCP and Odoo: a logical match
The MCP ecosystem in 2026
FAQ

What is the Model Context Protocol (MCP)?

The Model Context Protocol is an open communication protocol published by Anthropic on November 25, 2024. It defines a standard way, readable by any compatible AI, to describe tools and expose data to conversational assistants such as Claude, ChatGPT, or Cursor.

Let's use a simple analogy. Before USB-C, each phone, monitor or accessory had its own connector. To link device A to device B, you needed a dedicated cable. MCP plays that exact role for AI: it's the universal connector between an assistant and your business tools.

In practice, MCP standardizes three things:

Tools the AI can invoke, for example "read an invoice", "create a quote", "extract a report".
Resources the AI can read: a document, a database, a file.
Prompts the server can suggest to the user as predefined templates.

The full specification is public and documented on modelcontextprotocol.io. It relies on JSON-RPC 2.0, a battle-tested message format, and works both locally (on your workstation) and remotely (over a server).

💡 Key point: MCP is not a product, an AI, or a paid service. It is a technical standard comparable to HTTP or REST. Anyone can publish an MCP server, and any compatible AI assistant can connect to it.

Why was MCP created?

Picture the situation before MCP. You want Claude to read your emails, ChatGPT to query your CRM, and Cursor to run queries on your database. Each AI vendor ships its own "plugin" system with its own conventions, its own authentication, its own logic. The result: for each data source, you must build as many integrations as there are AI assistants.

This is the well-known N × M problem. With N AI assistants and M business tools, you have to maintain N × M integrations. Anthropic argued this was unsustainable at scale.

MCP turns N × M into N + M. Each AI implements the standard once (on the client side), each tool implements it once (on the server side), and every combination becomes possible with no additional effort.

The inspiration claimed by Anthropic is explicit: MCP aims to play the same role as the Language Server Protocol (LSP) for code editors. Before LSP, every IDE had to code its own support for every language. After LSP, any language could plug into any IDE by following the specification. MCP follows that proven pattern.

Without MCP	With MCP
Vendor-specific integrations	One open standard
N × M connectors to maintain	N + M servers/clients
No reuse across AIs	Any server works with any assistant
Ad hoc authentication	Standardized OAuth 2.1 (since 2025)
Scattered documentation	Single public specification

How MCP works: client-server architecture

MCP relies on three actors: a host, a client, a server.

The host is the application that contains the AI: Claude Desktop, ChatGPT Desktop, Cursor, or your own app.
The MCP client is the component inside the host that speaks the protocol. It opens connections to servers.
The MCP server is the program that exposes your tools and data. It can run locally or on a remote server.

The dialogue happens in phases. First, initialization: the client connects, negotiates capabilities and the protocol version. Then discovery: the server lists its tools (tools/list), its resources (resources/list) and its prompts. Finally, execution: the AI asks to invoke a tool (tools/call), the server runs the request and returns a structured result.

The dialogue uses JSON-RPC 2.0, a readable and widely supported remote procedure call format. Messages travel over three possible transports: stdio (standard input/output, ideal for local servers), HTTP+SSE (Server-Sent Events), and now Streamable HTTP for remote deployments.

⚠️ What to avoid: exposing an MCP server on the open Internet with no authentication or encryption layer. The specification recommends OAuth 2.1 for remote servers. Always verify the origin of an MCP server before connecting it to an assistant that has access to your sensitive data.

For more on the internals, the Anthropic documentation on MCP and the official GitHub repository gather specifications, SDKs (Python, TypeScript, Java, C#, Swift…) and ready-to-use server examples.

The main MCP use cases

MCP is not a lab curiosity. Within months, hundreds of public servers have been published. Here are the families of use cases most useful for a business.

1. Connecting an ERP like Odoo to an AI assistant

This is the case we care about most. Odoo is an open-source ERP covering accounting, sales, purchasing, inventory, HR, manufacturing… Its functional depth is huge, but so is its learning curve. With an MCP server exposing Odoo, you ask in plain English: "List unpaid invoices over 30 days" or "Create a quote for Helios Industries", and the AI translates your request into calls to the Odoo External API.

2. Reading and writing to databases

An MCP server can expose PostgreSQL, MySQL, MongoDB. The AI runs read queries (and, with validation, writes) without you needing to know SQL. Ideal for business analysts who want to explore data without coding.

3. Automating everyday tools

Gmail, Google Drive, Notion, Slack, Trello, GitHub, Linear: all of these now have MCP servers. You can ask your assistant to draft a reply to a customer, file documents, or create a GitHub issue out of a conversation.

4. Driving internal tools

IT teams can expose admin scripts, internal APIs or knowledge bases through an MCP server. The AI becomes a unified interface on top of your existing tools, with no rewrite.

The question that comes up most often when introducing MCP to an executive or a CIO is legitimate: "if I let an AI call my tools, where does my data go?". The specification anticipated this concern, and a serious connector extends it with safeguards.

First layer: scope is defined by you. An MCP server only does what its developer has explicitly coded. If you only expose invoice reads, the AI cannot modify your accounting. The principle of least privilege applies naturally.

Second layer: writes are validated. A properly configured AI assistant asks for confirmation before any sensitive action (create, update, delete). Critical workflows can be represented as a graph approved step by step.

Third layer: calls are auditable. Every tool invocation can be logged on the server with timestamp, user, parameters and result. This traceability is essential to meet GDPR and EU AI Act requirements, which are coming into force progressively through 2027.

Fourth layer: your data stays in your system. A well-built MCP server is a connector, not a copy. The data stays in Odoo, the MCP server exposes it on demand, the AI consumes it for the duration of the conversation. No permanent third-party storage is required.

Want to dig deeper? The Aidoo team can walk you through its approach to security and data sovereignty for Odoo-MCP connectors. Let's talk about your project or book a demo.

MCP and Odoo: a logical match

Why does Odoo benefit so much from MCP? Because Odoo is both deeply featured and deeply technical. Its 80+ apps cover every aspect of running a business, but mastering them takes weeks of training. Yet the business user just wants an answer: their margin rate, their late deliveries, their revenue by product.

Imagine the following situation. You run a professional supplies store and handle 200 quotes a month in Odoo. Without MCP, to chase quotes pending for more than 15 days you have to: open the Sales module, build a filter, export to CSV, open a spreadsheet, format it, copy emails, open your mailbox, draft 30 follow-ups. With an MCP server, you ask Claude: "Which quotes have been pending for more than 15 days? Draft a personalized follow-up for each." Three minutes later, drafts are ready for review.

That is exactly the mission of Aidoo AI: turning Odoo into a conversational interface while respecting your instance's roles, permissions and traceability. Browse our client use cases for real deployments, or check the pricing page which offers a 14-day free trial. If you are still on an older version, see our Odoo 19 migration service.

The MCP ecosystem in 2026

MCP adoption has been one of the fastest in recent enterprise software. Within months, the major AI players have announced support. The tech press, from TechCrunch to ZDNet, regularly covers new MCP servers.

On the tooling side, momentum is just as visible: public MCP server registries, dedicated sessions at developer conferences, official SDKs for most modern languages. The logic is the one of any open standard that takes off: more clients increase the value of new servers; more servers increase the incentive for clients to align.

For a company choosing its AI integration strategy today, MCP is a reasonable bet. You're not betting on a single vendor, you're betting on a protocol, exactly as people bet on HTTP in the nineties.

For a full overview of the topic (Odoo edition, migration, security, connecting Claude/ChatGPT), see our pillar page Odoo + AI integration.

FAQ

Is the Model Context Protocol free?

Yes. MCP is an open standard published under a permissive license. You pay no royalty to Anthropic to use it. Only the third-party services you connect (a Claude subscription, a cloud server, a commercial connector like Aidoo) may carry a cost.

Do I need to be a developer to use MCP?

No, not to use it as an end user. If you use Claude Desktop with a preinstalled MCP server, you just ask your questions. However, building an MCP server requires development skills (Python, TypeScript, or another supported language).

Does MCP work with ChatGPT and other AIs?

Yes. Although created by Anthropic, MCP is designed as a universal standard. OpenAI has announced support for ChatGPT and its Agents SDK. Other clients such as Cursor, Cline, Continue, Zed and several IDEs also integrate it.

What's the difference between MCP and a classic REST API?

A REST API is designed to be called by developers who know its structure. An MCP server is designed to be self-describing: it tells the AI which tools it offers, what they do and how to invoke them. The AI can therefore use it without prior human documentation.

MCP itself is a protocol, so it is GDPR-neutral. What matters is how your MCP server is configured: where it runs, who accesses it, what logs are kept, and whether data leaves your instance. A connector designed for GDPR compliance (self-hosting, encrypted data, logging, Odoo-role scoping) is entirely feasible.

Can I test MCP without commitment?

Yes. You can install Claude Desktop, add a local MCP server (for example the SQLite database sample provided by Anthropic) and experiment in minutes. For a test on Odoo, book an Aidoo demo or contact the team.

Conclusion: a standard that goes beyond hype

The Model Context Protocol is not a passing trend, it is infrastructure. It joins the long list of open standards that have shaped digital: HTTP, REST, OAuth, LSP. By giving AI and business tools a common language, MCP unlocks an experience that ERP users like Odoo's have been waiting for years: ask your software a question and get an answer.

For Odoo teams, this is an opportunity to reduce dependence on technical experts, accelerate onboarding for new users, and make the ERP useful to a much broader audience, all while keeping control of your data.

🚀 Have an Odoo project, or want to bring AI to your ERP?

The Aidoo team supports SMBs and mid-market companies on MCP-Odoo integration, from scoping to production. Free demo, quote in 48 hours.

Contact the Aidoo team →