# MCP Server An **MCP server** is a service that exposes tools, resources, and prompts to AI agents via the [Model Context Protocol](mcp). Servers are the capability provider side of the MCP architecture — they define what actions an agent can take and respond to tool call requests from [MCP clients](mcp-client). ## What an MCP Server Provides MCP servers expose three types of capabilities: - **[Tools](mcp-tool)** — Callable functions that the agent can invoke to take actions (e.g., `createTask`, `reply`, `search`, `readFile`) - **Resources** — Data sources the agent can read (e.g., database records, file contents, API responses) - **Prompts** — Reusable prompt templates that provide context or instructions ## AgentRQ as an MCP Server AgentRQ is a hosted MCP server designed for [human-in-the-loop](human-in-the-loop) agent workflows. It exposes tools that allow [Claude Code](claude-code) agents to communicate with their human principals in real time: | Tool | Description | |------|-------------| | `createTask` | Create a task on the task board | | `reply` | Send a message to the human | | `getTaskMessages` | Retrieve conversation history | | `updateTaskStatus` | Update a task's status | | `downloadAttachment` | Download a file attached to a task | | `getWorkspace` | Get workspace context | ## MCP Server Transports | Transport | Use Case | |-----------|----------| | stdio | Local servers running on the same machine | | HTTP + SSE | Remote servers accessible over the network | AgentRQ uses HTTP + [SSE](sse) transport, which means it can be accessed from any machine where Claude Code runs. ## Building Your Own MCP Server The MCP specification is open. You can build custom MCP servers in any language to expose your own APIs, databases, or services to AI agents. The Anthropic MCP SDK is available for TypeScript, Python, and other languages. ## Related Terms - [MCP](mcp) - [MCP Client](mcp-client) - [MCP Tool](mcp-tool) - [SSE](sse) - [Tool Use](tool-use)