---
title: "MCP — Glossary | LLMind"
description: "Model Context Protocol — Anthropic's open standard for connecting AI agents to tools, files, and data sources."
url: https://llmind.org/glossary/mcp/
source_format: html
---
[← Glossary](https://llmind.org/glossary/)

# MCP

**Model Context Protocol — Anthropic's open standard for connecting AI agents to tools, files, and data sources.**

MCP (Model Context Protocol) is an open protocol introduced by Anthropic in late 2024 that standardizes how AI agents and language models communicate with external tools, file systems, and data sources. Instead of each AI application implementing its own ad-hoc interface to tools and data, MCP provides a single, extensible protocol that tools and data-source providers can implement once and have work with any MCP-compatible client.

## What MCP is

MCP is a client-server protocol. An AI agent or language model is the client; it sends requests to MCP servers. A server exposes capabilities: file system access, database queries, API calls, computations. The protocol is message-based and JSON-serialized, making it language-agnostic and straightforward to implement. A single MCP server can run as a local subprocess, a network service, or embedded in an application.

## Filesystem MCP and file access

The most common use case is filesystem access. Anthropic's reference filesystem MCP server (part of the MCP ecosystem at https://modelcontextprotocol.io/) exposes a directory to an agent with standard operations: list files, read files, write files, and search. Pair this with LRFS-enriched files, and the agent can read structured metadata alongside raw bytes — the agent understands content context without re-parsing.

## Why MCP matters for LLMind

File access is a primary agent use case. MCP makes the transport and tool interface standard; LLMind makes the files themselves richer. When an MCP filesystem server serves a directory of LLMind-enriched files, agents can leverage semantic metadata directly. The two initiatives are complementary — MCP provides the plumbing; LRFS provides the signal.

## Related terms

-   [Embedded metadata](https://llmind.org/glossary/embedded-metadata/)
-   [LLM-ready file](https://llmind.org/glossary/llm-ready-file/)

## See also

-   [MCP integration hub](https://llmind.org/mcp/)
-   [Claude Desktop guide](https://llmind.org/mcp/claude-desktop/)
-   [AI agents file access use case](https://llmind.org/use-cases/ai-agents-file-access/)
-   [Learn](https://llmind.org/learn/)
-   [Spec](https://llmind.org/spec/)