---
title: "MCP (Model Context Protocol) | AgentRQ Glossary"
url: https://agentrq.com/glossary/mcp
description: "MCP (Model Context Protocol) is an open protocol by Anthropic that standardizes how AI agents connect to external tools, data sources, and services using a client-server architecture."
---

[Back to Glossary](https://agentrq.com/glossary/)

# MCP (Model Context Protocol)

**MCP (Model Context Protocol)** is an open protocol developed by Anthropic that standardizes how AI agents connect to external tools, data sources, and services. MCP defines a common language that [MCP clients](https://agentrq.com/glossary/mcp-client) (like [Claude Code](https://agentrq.com/glossary/claude-code)) use to communicate with [MCP servers](https://agentrq.com/glossary/mcp-server) (like AgentRQ) — enabling a rich ecosystem of interoperable agent capabilities.

## Why MCP Exists

Before MCP, every AI tool had to build custom integrations for every service it wanted to connect to. This created a fragmented ecosystem where each LLM provider and each tool had incompatible APIs.

MCP solves this with a standard protocol. A server built to the MCP spec works with any MCP-compatible client, and a client built to the MCP spec can use any MCP server — without custom code.

## MCP Architecture

text

```text
AI Agent (MCP Client)
        │
        │  MCP Protocol (JSON-RPC over stdio/SSE/HTTP)
        ▼
  MCP Server (e.g., AgentRQ)
        │
        ▼
  External Service / Tool
```

The protocol has three primitives:

- → **[Tools](https://agentrq.com/glossary/mcp-tool)** — Functions the agent can call (e.g., AgentRQ's `task` and `memory`)
- → **Resources** — Data sources the agent can read (e.g., files, databases)
- → **Prompts** — Reusable prompt templates the server exposes

## MCP and AgentRQ

AgentRQ is an MCP server. When you add AgentRQ to your [Claude Code](https://agentrq.com/glossary/claude-code) configuration, Claude Code becomes an [MCP client](https://agentrq.com/glossary/mcp-client) that can call AgentRQ's tools:

- → `task` — Create, get and update tasks on your [task board](https://agentrq.com/glossary/task-board), reply to the human or ask them a question, and fetch attachments
- → `memory` — Load, save and delete notes that outlive a task
- → `skills` — Search, load, save and delete the workspace's skills
- → `workspace` — Get workspace configuration and context, and publish events
- → `siteTools` — Search and call the tools of websites shared from Chrome
- → `pano` — Shared markdown docs across the account's workspaces
- → `help` — Show every tool's actions, and each action's payload with an example

Each tool takes an `action` and a `payload`, for example `task {"action": "reply", "payload": {"taskId": "…", "text": "Done!"}}`. The older one-tool-per-action names (`createTask`, `getTask`, `updateTaskStatus` and so on) still work for clients that are already installed.

## MCP Transport Options

| Transport | Description |
| --- | --- |
| stdio | Local server, communicates over stdin/stdout |
| HTTP + SSE | Remote server, uses HTTP for requests and [SSE](https://agentrq.com/glossary/sse) for streaming |

AgentRQ uses HTTP + SSE transport, enabling real-time [notifications](https://agentrq.com/glossary/notification) from anywhere.

## Related Terms

- → [MCP Client](https://agentrq.com/glossary/mcp-client)
- → [MCP Server](https://agentrq.com/glossary/mcp-server)
- → [MCP Tool](https://agentrq.com/glossary/mcp-tool)
- → [SSE](https://agentrq.com/glossary/sse)
- → [Claude Code](https://agentrq.com/glossary/claude-code)