Last reviewed:
What is MCP (Model Context Protocol)? Definition and business implications
MCP (Model Context Protocol) is an open standard, introduced by Anthropic in November 2024, that lets an AI model connect to data sources and external tools in a uniform way. It avoids writing specific connectors for every model-application combination.
Before MCP, every company deploying an AI assistant had to code custom connectors: one for its CRM, one for its document base, one for its internal tools. Worse, these connectors had to be redone with each model change. MCP standardises this integration. It defines a client-server architecture: the LLM is the client, external tools (databases, business APIs, files) are exposed by MCP servers. An MCP server written once is usable by all compatible models. The protocol is open source. In December 2025, Anthropic handed it over to the Agentic AI Foundation, a foundation hosted by the Linux Foundation and co-founded with OpenAI and Block; the current specification is dated 28 July 2026. MCP is supported by ChatGPT, Claude, Gemini, Microsoft Copilot, Cursor and Visual Studio Code, among others. By the end of 2025, there were more than 10,000 active MCP servers, covering Slack, GitHub, Salesforce, Google Drive, and most standard enterprise SaaS tools.
Concrete example
Illustrative case. A 40-employee accounting firm wants its AI assistant to query directly the accounting software (Sage, Cegid), the internal document management system, and the shared mailbox. Without MCP, that means three specific connectors to develop, with development and ongoing maintenance at its own expense. With MCP, if an MCP server exists for each tool, it is enough to plug them in, and maintenance falls to the server publishers. Better: if the firm decides to migrate from Claude to another model, no rewriting on the connector side, the standard protocol remains.
FAQ
What is MCP (Model Context Protocol)?
It is an open standard, introduced by Anthropic in November 2024, that lets an AI model connect to data sources and external tools in a uniform way. It avoids writing a specific connector for every model-application combination.
What problem does MCP actually solve?
Before it, every company had to code custom connectors (one for the CRM, one for the document base, and so on), to be redone with each model change. MCP standardises this integration through a client-server architecture: an MCP server written once is usable by all compatible models. By the end of 2025 there were more than 10,000 active MCP servers covering most enterprise SaaS tools.
What does MCP change for our integration costs?
It can bring them down. Where each internal tool required a custom connector, MCP lets you reuse an existing server, which cuts development and maintenance effort. The size of the saving depends on whether an MCP server exists for each of your tools.
Does MCP reduce our dependence on an AI provider?
Yes. Because the protocol is standard, switching model provider no longer requires rewriting everything on the connector side. This real interoperability reduces technical dependence on a single actor and makes multi-provider arbitration easier.
Does MCP create a new security risk?
Yes. MCP servers installed on an employee's machine open access to sensitive tools. The map of authorised MCP servers therefore becomes a governance topic in its own right, on a par with browser extensions.
See also
Further reading
Sources
- Introducing the Model Context Protocol, Anthropic, November 2024. https://www.anthropic.com/news/model-context-protocol
- Model Context Protocol specification, version 2026-07-28. https://modelcontextprotocol.io/specification
- MCP joins the Agentic AI Foundation, official Model Context Protocol blog, 9 December 2025. https://blog.modelcontextprotocol.io/posts/2025-12-09-mcp-joins-agentic-ai-foundation/
- Linux Foundation Announces the Formation of the Agentic AI Foundation (AAIF), press release, December 2025. https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation