MCP
also called Model Context Protocol
An open protocol for plugging tools and data into any agent.
A standard way to expose tools, resources and prompts to AI applications. An MCP server describes its capabilities once; any MCP-compatible client can use them.
Before MCP, every assistant needed a bespoke integration per tool. MCP turns integrations into a marketplace: write one server for your internal system and every compatible agent can use it.
The value of a standard is measured in what does not have to be rewritten. Before MCP, connecting an agent to ten systems meant ten bespoke integrations, each maintained against one vendor. A protocol turns that into ten servers that any compliant client can use — and its move to neutral foundation stewardship is what makes it reasonable to bet on.
Write the server once
Exposing an internal ticketing system to whatever agent the team uses this quarter.
Build one MCP server. Today’s harness connects to it, next quarter’s connects to it too, and the alternative — a bespoke integration per tool per vendor — is the maintenance burden the protocol exists to remove.
The trust boundary you just created
Connecting a third-party MCP server to an agent with filesystem access.
That server can return content that steers the agent — the specification’s own security section says so. Treat remote servers as untrusted input sources: require approval for consequential actions and scope credentials to the minimum that works.
- standardModel Context Protocol specification ↗
The normative document: hosts, clients, servers, JSON-RPC, resources, prompts and tools — plus the security section on consent, user control and care around tool execution.
- blogIntroducing the Model Context Protocol ↗
The original MCP framing: an open standard that replaces per-integration glue with a client/server contract for tools and data.
- blogDonating the Model Context Protocol and establishing the Agentic AI Foundation ↗
MCP moves to vendor-neutral foundation stewardship — the reason it is safe to treat MCP as infrastructure rather than one vendor’s feature.
- docsMCP servers ↗
Connecting a model to remote or tunnelled private MCP servers, plus the warning that an untrusted server can pull sensitive data into context.
Tool
A function the model is allowed to call to affect the world.
Function Calling · Tool use
The model-provider feature that makes structured tool calls reliable.
Permission Request
The agent pausing to ask before doing something risky.
Harness · Scaffold
Everything around the model that makes it an agent.