What is MCP?
Last reviewed 7 October 2026
The Model Context Protocol (MCP) is an open protocol that gives AI applications a standard way to connect to external tools, data sources and prompts, so one integration can work with any compatible client. Anthropic announced it in November 2024. It uses JSON-RPC 2.0 messages between hosts, clients and servers, and takes some inspiration from the Language Server Protocol.
Also called: Model Context Protocol, MCP protocol, MCP standard.
Why Model Context Protocol (MCP) matters
Before MCP, each AI application needed its own connector for each system. Anthropic’s announcement described the problem directly: every new data source required its own custom implementation. MCP makes a tool something you build once as a server and use from any MCP-capable client, much as the Language Server Protocol let one language server work across many code editors.
It also concentrates risk. Connecting an agent to an MCP server gives it whatever that server can do. The specification is plain that tools represent arbitrary code execution, that tool descriptions should be treated as untrusted unless they come from a trusted server, and that hosts must obtain explicit user consent before invoking any tool. It also says the protocol cannot enforce these principles itself; implementers have to.
How it works
MCP has three roles. A host is the AI application the user works in. Inside it, a client is a connector that talks to a server. A server is a service that provides context and capabilities. Servers can offer resources (data for the user or the model), prompts (templated messages and workflows) and tools (functions the model can execute). Clients can offer servers elicitation, a way to ask the user for more information.
At run time the host connects to its configured servers, learns which tools each offers, and passes those tool descriptions to the model. When the model asks to use one, the client sends the call to the server and returns the result. To the model this looks like ordinary function calling; MCP standardises what happens on the other side.
The specification is versioned by date. The version we read on 7 October 2026 is dated 28 July 2026 and also defines optional extensions, such as tasks for long-running operations, which a client and server use only if both support them.
Example: one ticketing server, two clients
An engineering team writes an MCP server for its internal ticketing system with three tools: search tickets, read a ticket and add a comment. A developer connects it to a coding assistant, and an operations lead connects the same server to a desktop AI app.
Both clients see the same three tools without any client-specific code. The team decides that adding a comment needs confirmation, so both hosts ask the user before running it. When the team adds a fourth tool, both clients see it the next time they connect.
How Swfte relates to it
Built in the product
Swfte has an MCP gateway in its runtime. It checks that a caller’s API key is scoped to the MCP server and belongs to the workspace, and logs each call with its method, status and latency. Per-tool allow and deny rules, per-call rate limits and SIEM export are not built. Agents built in Swfte Studio can act as MCP clients, and Swfte Cortex can use external MCP servers with an approval gate on destructive tools. Swfte does not expose your agents as an MCP server.
This page is the short definition. The MCP gateway page covers the controls in detail and states what Swfte has and has not built.
- MCP gateway: What it should control and what Swfte has built
- Swfte Studio
- Swfte Cortex
Related terms
- MCP server
An MCP server is a program that exposes capabilities to AI applications through the Model Context Protocol, so a model can call its tools, read its data and use its prompt templates.
- Function calling
Function calling is a feature of language model APIs that lets a model return a structured request to run a function you defined, with arguments that match its schema, instead of only replying in text.
- AI agent
An AI agent is a software program that uses an AI model to decide what to do next and then acts through tools to reach a goal it was given.
- Human-in-the-loop (HITL)
Human-in-the-loop (HITL) refers to a system in which a person takes an active part in the operation, supervision or decisions of an automated process at defined points.
Common questions
- Who created MCP?
- Anthropic introduced the Model Context Protocol in November 2024, releasing the specification, SDKs, support for local MCP servers in Claude Desktop and a repository of open-source servers. The protocol is open, and the specification and documentation are published at modelcontextprotocol.io.
- Is MCP the same as an API?
- No. An API is the interface a system exposes, in whatever shape its owner chose. MCP is a standard wrapper that describes tools, data and prompts in one consistent way for AI applications. An MCP server usually calls the system’s existing API behind the scenes, so MCP adds a common layer rather than replacing APIs.
- Is MCP secure?
- The protocol states consent and safety principles but cannot enforce them; the host, client and server implementations do. Risks include over-broad credentials, untrusted tool descriptions and prompt injection through tool results. Use servers you trust, scope their credentials narrowly, require confirmation for actions with side effects and log every call.
- Do I need an MCP gateway?
- Not for one person trying a few servers. Once many agents or people use many servers, a gateway gives one place to hold credentials, decide access, log calls and limit usage. Check what a given gateway actually enforces, because coverage differs a lot between products.
Sources
Definitions on this page were read on the sources below on 7 October 2026. Where sources define the term differently, the page says so. The full glossary lists more terms.
- Model Context Protocol specification, current version (read 2026-10-07)
- Anthropic announcement introducing the Model Context Protocol (read 2026-10-07)