Model Context Protocol

How MCP works and why Agent Handler uses it.

Model Context Protocol (MCP) is an open standard from Anthropic that lets AI models read from and write to external systems through a consistent interface. See the official MCP docs for the full specification.

How MCP is structured

MCP defines three components:

  • MCP client — runs inside the LLM application and initiates tool calls
  • MCP server — exposes data and actions from an external system to clients
  • Tools — the individual actions or queries an MCP server exposes (for example, create_issue or search_contacts)
Diagram showing an MCP client in an LLM application communicating with an MCP server that wraps an external system

Related: How MCP compares to APIs

Why MCP matters

  • One protocol per integration. LLM providers and SaaS apps implement MCP once instead of maintaining a bespoke integration per client.
  • Structured context for agents. MCP gives agents a consistent way to retrieve, update, and pass context, which is required for multi-step workflows.
  • Lower LLM overhead. Standardized context formats reduce the tokens and processing the model spends interpreting tool schemas.
  • Centralized security posture. Governance over how context is stored, shared, and updated lives in one layer rather than per integration.

Related: Tips for using MCP