MCP standardizes an integration boundary. An AI application acts as a host; its clients communicate with servers that expose supported capabilities. The underlying model need not speak MCP itself: the application translates between model behavior and protocol operations.
A shared interface reduces repeated integration work, but “MCP-compatible” does not mean every client supports every server feature. Check the protocol version, transport, and capabilities of the implementations you plan to connect.
Example: a document server can expose search and reads to several AI applications. Whether an agent may read a particular file or call a write tool still depends on authentication, authorization, and the host’s controls. A standard connector does not confer unrestricted access.
Angie Jones describes Block’s MCP-based rollout in episode 48. For the implementation decision, see MCP vs. custom tool integration and agent identity and permissions. Protocol details here were checked against the current official overview in October 2026.
See AI vocabulary in plain English for this term alongside the other language of models, data, agents, and infrastructure.
Sources
- Model Context Protocol: Architecture overview — Official host/client/server model and exposed primitives; checked October 2026.
Go deeper
- What can MCP actually do? AI, decoded · 3 Things MCP Unlocks That a Chatbot Can't
- Should you use MCP or build a custom integration to connect AI to your tools? AI, decoded · MCP vs. Custom Integration: How to Connect AI to Your Tools
- Do you have to centralize your data before AI agents can use it? AI, decoded · Centralizing Data Before AI: Why the Lake Doesn't Save You
- MCP: Tools specification (2026-07-28) docs
Follow tools/list and tools/call to see how a server describes tools and returns results.
- MCP: Authorization specification (2026-07-28) docs
Study authorization for HTTP transports, including scopes and token audience checks. Stdio credentials follow a separate path.
From the conversation
-
How Block Deployed AI Agents to 12,000 Employees in 8 Weeks w/ MCP | Angie Jones -
Context Poisoning is Killing Your AI Agents: How to Stop it | Michel Tricot, Airbyte -
Why Context Alone Isn't Enough for Enterprise AI Agents | WisdomAI CPO Kapil Chhabra -
Architecting Reliable Agentic AI | Cisco’s Giovanna Carofiglio on the AGNTCY Collective -
Data Federation, Not Centralization, Is What Enterprise AI Needs | Jitender Aswani, Starburst -
What Breaks When Agents Become the Customer | Svix CEO Tom Hacohen -
I Built an AI Coworker That Runs 90% of My Day | Sterling Chin, Postman -
From Demo to Defensibility: How to Build an AI Business that Lasts | Aurimas Griciūnas