What is the Model Context Protocol (MCP)?
The Model Context Protocol (MCP) is an open protocol, introduced by Anthropic in November 2024, that standardises how AI applications connect to external tools and data through servers that expose tools, resources and prompts.
Checked
Before MCP, every AI application wrote its own connector for every system. MCP defines one protocol instead: a system is wrapped once as an MCP server, and any compatible application can use it. The specification says it takes inspiration from the Language Server Protocol, which did the same for programming languages and code editors.
Anthropic open-sourced MCP on 2024-11-25 with a specification, SDKs and a repository of servers. On 2025-12-09 it became a founding project of the Linux Foundation's Agentic AI Foundation, alongside AGENTS.md and Block's goose.
How it works
Hosts, clients and servers
The host is the AI application, a client inside it holds the connection, and a server provides context and capabilities.
Messages
Communication uses JSON-RPC 2.0 messages, with capabilities negotiated per request.
Server features
Tools are functions the model can execute, resources are context and data, and prompts are templated messages and workflows for users.
Client features
Elicitation lets a server ask the user for more information through the host.
Where the specification stands
The current specification version is dated 2026-07-28. It describes stateless, self-contained requests and a set of opt-in extensions, including Tasks for long-running operations, MCP Apps for interactive interface elements, and Skills over MCP for structured workflow instructions.
Context cost and security
Loading every tool definition from every server up front spends context the agent needs for the task. Anthropic's code execution with MCP approach, which loads tools on demand and filters data in a sandbox, took one example from about 150,000 tokens to about 2,000.
The specification is direct about risk: tools represent arbitrary code execution, tool descriptions should be treated as untrusted unless the server is trusted, and hosts must obtain user consent before invoking a tool. Any MCP server's output is also a route for prompt injection.
How Motiontic's engineers apply it
Motiontic's engineers wrap a client system as an MCP server when more than one agent or application will need it, keep each server's tool set small, and scope credentials to the narrowest permissions the task needs. Third-party servers are reviewed like any dependency before an agent can call them.
The three MCP server features
| Feature | What it is | Used for |
|---|---|---|
| Tools | Functions the model can execute | Actions and lookups: create a ticket, query a database |
| Resources | Context and data for the user or the model | Files, records or documents to read |
| Prompts | Templated messages and workflows | Reusable instructions a user can pick |
Common questions
Who created MCP?
Anthropic introduced and open-sourced MCP on 2024-11-25. It is now a founding project of the Linux Foundation's Agentic AI Foundation, announced on 2025-12-09.
What is an MCP server?
An MCP server is a service that exposes tools, resources or prompts over the Model Context Protocol, so any compatible AI application can use them without a custom connector.
How is MCP different from A2A?
MCP connects an agent to tools and data. A2A connects agents to each other so one can discover another and hand it a task. Most systems that use A2A also use MCP.
Is MCP secure?
The protocol cannot enforce security on its own. The specification treats tools as arbitrary code execution and requires user consent before tool calls, so safety depends on how hosts and servers implement permissions, credentials and review.
Related
MCP vs A2A
The protocol for tools and data, compared with the protocol for agents talking to agents.
Skills vs MCP
Connectivity to systems, compared with packaged expertise the agent loads on demand.
A2A protocol
The open protocol for agents talking to other agents, now under the Linux Foundation.
Tool calling
How a model asks software to do something: a named tool, typed arguments, a result.
AI agent development
Agents that do real work for users, with the evals, traces and guardrails to keep them in production.
Running agents that have to hold up?
Describe the agent, the systems it touches and what it must never get wrong. The first call is with a lab engineer, who will say plainly whether it is ready for production, and whether Motiontic is the right team.