--- title: "MCP" description: "An open protocol for connecting AI applications to sources of information and tools." category: "AI and tooling" canonical: "https://contentlexicon.com/mcp" updated: "2026-10-12" --- # MCP _AI and tooling_ > An open protocol for connecting AI applications to sources of information and tools. MCP is an open standard that defines how AI applications connect to outside systems such as document stores, [databases](https://contentlexicon.com/database) and business tools. Instead of building a bespoke integration for every pairing of AI tool and system, developers build to one shared specification, and anything that follows it can connect. Anthropic introduced it in November 2024, and it has since been adopted more widely, including by OpenAI and Google. The parts have plain jobs. An MCP client is the part of an AI application, such as a chat assistant or coding tool, that makes the connection. An MCP server is a small program that sits in front of a system and describes what it offers. Servers can offer tools, which are actions the model can take, such as searching records or creating a draft. They can offer resources, which are information the model can read, such as files or a [style guide](https://contentlexicon.com/style-guide). And they can offer [prompts](https://contentlexicon.com/prompt), which are reusable templates for common tasks. The protocol standardises how all of these are described and used. It doesn't decide what a system contains or whether it's trustworthy. ## Why it matters for content MCP makes live content systems available to AI assistants while they work. A style guide, terminology list, [knowledge base](https://contentlexicon.com/knowledge-base) or product catalogue can become a source the assistant consults during drafting, rather than a document pasted into a prompt and gradually forgotten. When the style guide changes, every connected assistant sees the new version. That raises the value of well-structured, authoritative content. A team could let an assistant look up the current style rule, check a draft against it and prepare an amendment in the same workspace. The benefit comes from maintained sources and well-defined actions, not from the connection alone. Writing is part of the design. Each tool and resource has a name and description that the model reads to decide when to use it. Vague descriptions lead to wrong choices, so content people should help write and test them. Permissions need the same care. A connection exposes whatever its server provides, within the access the application and server enforce. A read-only connection to a published style guide carries little risk. One that can edit or publish needs tighter limits, logging and human confirmation for consequential actions. The protocol supports these controls but doesn't enforce any organisation's policy. ## How MCP changes the picture MCP moves AI from isolated chat towards agents that work inside organisational systems, finding information and taking actions across several tools in one task. The more an agent can do, the more its sources and permissions matter. Content an agent reads can also carry instructions. A web page, [email](https://contentlexicon.com/email) or document containing hidden text such as "ignore previous instructions" can steer an agent that has access to other tools, a risk known as prompt injection. Teams need to test how tools are selected, how errors are handled, where answers came from and what happens when a source is hostile. The protocol supplies a route, but it doesn't make every use of that route safe. ## Questions people ask **Is MCP a database?** No. It's a protocol that lets AI applications connect to systems that may contain data or actions. **Does MCP give an AI access to everything?** No. A connection exposes the capabilities its server provides, within the access the application and server enforce. Grant only what's needed, and don't treat the protocol name as assurance that every permission boundary is correct. **Is MCP only for developers?** Building a server is technical, but content people should shape source quality, tool descriptions, permissions and review. **Does MCP replace APIs?** No. An MCP server often sits on top of an existing [API](https://contentlexicon.com/api) and presents its capabilities in a consistent form for AI applications. ## Learn more - [What is MCP?](https://modelcontextprotocol.io/docs/2026-07-28/getting-started/intro) – the official introduction under the 2026-07-28 documentation version, explaining the connection model; version context matters for technical details. - [MCP security best practices](https://modelcontextprotocol.io/docs/2026-07-28/tutorials/security/security_best_practices) – the more consequential reading on consent, token handling, scope and attack surfaces. - [Introducing the Model Context Protocol](https://www.anthropic.com/news/model-context-protocol) – the original announcement and the problem the standard was created to solve. Related terms: [Agentic AI](https://contentlexicon.com/agentic-ai), [API](https://contentlexicon.com/api), [Knowledge base](https://contentlexicon.com/knowledge-base), [Machine-readable](https://contentlexicon.com/machine-readable) Also called: Model Context Protocol This page was last reviewed in October 2026.