Model Context Protocol (MCP)
An open standard for connecting AI agents to external data sources, tools, and services through a consistent interface. The 'USB-C for AI integrations.'
Also known as: MCP · Model Context Protocol · AI agent protocol
What is Model Context Protocol (MCP)?
Model Context Protocol (MCP) is an open protocol introduced by Anthropic that standardizes how AI agents connect to external data sources, tools and services. Instead of custom integrations, any MCP-compatible agent can call any MCP server through a shared interface, so a server built once, such as one for Gmail, works for every agent that needs it.
Christos Papadimitriou, theagency47 · Updated May 18, 2026Why MCP matters
The pre-MCP world had a quiet but expensive problem: every agent-to-tool integration was custom. A custom AI agent that needed to read Gmail, write to Slack, and update HubSpot was three bespoke integrations, and if the client wanted to swap their underlying model later, all three had to be reviewed. The integrations were the most fragile and least valuable line item on every build.
MCP collapses that. Once a vendor (Anthropic, OpenAI, Google, Microsoft, AWS, or any third party) ships an MCP server for a service, every MCP-compatible client can talk to that service through the shared protocol. A Gmail MCP server built once works for a Claude agent today and a GPT-5 agent next year. The integration is no longer the bottleneck.
How theagency47 uses MCP in agent builds
Most of our 2026 builds use MCP as the tool-use layer. For a Workforce Starter build that touches three integrations (CRM, calendar, email), we either use existing community MCP servers or write one new one, total integration time is hours, not days. The agent itself does not change when we swap one MCP server for another; the contract is at the protocol layer.
theagency47’s agents connect to company systems through MCP servers; see the AI Company Brain.
FAQ
Is MCP only for Claude?
No. MCP was introduced by Anthropic but is now an open standard with first-class support in OpenAI’s Agents SDK, Google’s Gemini, Microsoft Azure AI, and AWS Bedrock. Any MCP-compatible client can call any MCP server.
Do I need MCP if I have a small agent build?
Not strictly. A Spark build with one integration may be simpler with a direct API call. MCP earns its complexity at 3+ integrations or when the client expects to add more integrations over time.
What does an MCP server actually contain?
Three primitives: tools (functions the agent can call), resources (data the agent can read), and prompts (templated workflows). A Gmail MCP server, for instance, exposes tools like send_email and resources like list_inbox.
Related terms
Tool Use
An LLM's ability to call external functions or APIs as part of its response, enabling actions beyond text generation. Also called function calling.
Knowledge Base (RAG)
The collection of documents an AI agent retrieves from at decision time. The mechanism is called retrieval-augmented generation (RAG).
AI Agent
A software system that takes a goal, plans steps, calls tools, and acts autonomously to produce an output, without step-by-step human prompting.
