Model Context Protocol (MCP)
Model Context Protocol is an open standard from Anthropic that gives AI models a consistent way to call tools, fetch data, and act on external systems.
Model Context Protocol (MCP) is an open standard, originally introduced by Anthropic in late 2024, that lets AI assistants connect to external tools, data sources and services through a common interface. Where each AI vendor previously required bespoke integrations, MCP defines a single protocol that any model and any service can speak — making it the closest thing the AI ecosystem has to a "USB-C for AI tools".
Why MCP exists
An AI assistant becomes useful when it can read your documents, query your database, file a ticket in your system or call your API. Before MCP, every assistant–tool pairing was a one-off integration. MCP standardises the interface so a tool implemented once works across every compliant assistant — and an assistant that speaks MCP can use any compliant tool.
How MCP works
| Component | Role |
|---|---|
| MCP server | Wraps a tool or data source and exposes it over the MCP protocol |
| MCP client | Lives inside the AI assistant; discovers and calls available servers |
| Resources | Read-only data the model can fetch (documents, records, files) |
| Tools | Actions the model can invoke (create, update, run a query) |
| Prompts | Reusable templates a server can offer to the assistant |
An assistant can connect to many MCP servers at once — a filesystem server, a Postgres server, a GitHub server, an internal API server — and the model decides which to call to satisfy a request.
What MCP enables
- Portable tool ecosystems. Build a server once; use it from Claude, ChatGPT, Cursor, Zed and others.
- Local-first context. A user's own filesystem, database or notes can be made available to an assistant without uploading data to a vendor.
- Agentic workflows. Assistants can chain calls across multiple servers to accomplish multi-step tasks.
- Enterprise data access. Internal systems get a single protocol for AI integration rather than per-vendor adapters.
MCP and AI search
MCP is not a search-engine optimisation surface in the traditional sense — it does not directly determine whether your brand is cited in Google AI Overviews or ChatGPT Search. But it shapes the next layer: when AI assistants act as agents and need authoritative information about a brand, an MCP server can be the canonical machine-readable source. Forward-looking publishers and SaaS vendors are already shipping MCP servers alongside their public APIs.
MCP vs. plugins vs. function calling
| Approach | Scope | Portability |
|---|---|---|
| Vendor plugins | One vendor's assistant only | Locked in |
| Function calling | One application's API surface | Per-app integration |
| MCP | Any compliant assistant + any compliant server | Portable |
Frequently asked questions
Is MCP only for Claude?
No. Anthropic introduced MCP and ships first-class support, but the protocol is open. Multiple IDEs, agent frameworks and other assistants have adopted it.
Do I need MCP if I already have an API?
An API is for software clients. An MCP server is an opinionated wrapper that makes the same capability discoverable and callable by AI assistants without per-vendor work. The two coexist.
How does MCP relate to llms.txt?
llms.txt is a static, content-discovery file. MCP is a live, interactive protocol. Together they cover the static and dynamic ends of "how should an AI talk to my domain?"
Related Terms
Anthropic Claude
Claude is Anthropic's family of large language models. It is known for long-form reasoning, a careful tone, and the Constitutional AI training approach.
Grounding
Grounding is the practice of tying an AI model's answer to verifiable source material instead of letting it generate from memory alone.
Retrieval Augmented Generation (RAG)
Retrieval Augmented Generation lets an AI model fetch fresh information before it answers, instead of relying only on what it learned during training.
llms.txt
llms.txt is a proposed plain-text file at the root of a site. It gives large language models a curated, machine-readable map of the pages that matter most.
AI Agent
An AI agent is a model wired up to take actions on its own: read a brief, call tools, work through steps, and return a result without step-by-step prompting.
Large Language Model (LLM)
A large language model is an AI trained on huge amounts of text to predict the next token, which is enough to make it read, write, and reason in plain language.
