Skip to main content
GLOSSARY TERM

MCP (Model Context Protocol)

MCP (Model Context Protocol) is an open standard for connecting AI models and agents to data sources, services, and tools. It provides standardized interfaces for agents to access external context, execute actions, and integrate with third-party systems on the agentic web.

What MCP enables

  • Data source connections - Agents access databases, APIs, file systems, and knowledge bases through unified interfaces
  • Tool integration - Agents discover and use external tools and services programmatically
  • Context injection - Dynamic injection of relevant information into agent workflows
  • Action execution - Agents execute operations across multiple systems with standardized commands
  • Permission management - Fine-grained access control for agent capabilities

How MCP works

MCP defines three core primitives for agent-context interaction:

Resources

Expose data and content to agents (files, database records, API responses, documents)

Tools

Provide callable functions agents can execute (search, calculations, API calls, system commands)

Prompts

Define reusable prompt templates with dynamic context injection

MCP in the agentic web

MCP complements other agentic web protocols by focusing on context access while protocols like PEAC handle policy enforcement and A2A handles agent communication.

Example workflow combining MCP + PEAC:

  1. Agent discovers API through MCP resource registry
  2. Agent fetches /.well-known/peac.txt to understand pricing
  3. Agent pays using HTTP 402 flow
  4. Agent receives PEAC-Receipt as proof of payment
  5. Agent uses MCP to call API tools with receipt-based auth

MCP server architecture

MCP uses a client-server model where:

  • MCP Clients - AI applications and agents that consume context and tools
  • MCP Servers - Services that expose resources, tools, and prompts through standardized interfaces
  • MCP Specification - JSON-RPC based protocol for client-server communication

Use cases

  • Enterprise integrations - Connect AI agents to internal databases, CRMs, and business tools
  • Knowledge bases - Agents access company wikis, documentation, and institutional knowledge
  • API orchestration - Agents combine multiple third-party APIs into complex workflows
  • Tool marketplaces - Standardized discovery and usage of agent tools and plugins
  • Multi-modal data access - Unified interface for text, images, audio, and structured data

MCP and receipts

When MCP servers expose paid resources or tools, they can integrate with PEAC-Receipt verification. Agents present receipts when calling MCP tools, proving they have paid for access. This enables pay-per-use models for MCP resources while maintaining the standardized MCP interface.

Learn more