Model Context Protocol
An open standard that lets an AI application connect to external tools and data through one client/server interface instead of a bespoke integration per tool.
Last verified August 21, 2026 · 3 sources
Before MCP, connecting an assistant to Jira meant writing a Jira integration; connecting it to Google Drive meant writing a Drive integration. MCP replaces N integrations per application with one protocol: a server exposes tools and data, a client speaks the protocol, and any client can use any server.
Anthropic published it as an open standard, and it is now the ordinary way Claude reaches anything outside itself.
The two halves
| Role | What it is | Examples |
|---|---|---|
| MCP server | Exposes tools, resources and prompts over the protocol | A Jira server, a Postgres server, your internal API |
| MCP client | Consumes them | Claude Code, the Claude desktop app, an Agent SDK application |
You mostly install servers and rarely write clients.
What it does not do
- It is not a permission system. MCP describes what a tool can do, not what Claude may do. That is permissions, configured separately — and this distinction is the single most important thing to get right before connecting a server that can write.
- It is not a Skill. A Skill is instructions telling Claude how to do something; an MCP server is a connection giving Claude access to something. See Skills vs MCP.
- It is not a Connector. Connectors are the managed, end-user-facing integrations in the Claude apps. Many are MCP underneath, but the install path, the audience and the admin controls differ. See MCP vs Connectors.
- It is not automatically safe. A server runs code with whatever credentials you give it. Read the source of any community server before connecting it to a system that matters.
Organisational control
Anthropic documents managed MCP for controlling which servers an organisation’s users may connect to. If you are rolling Claude Code out across a company, that page matters more than the quickstart: without it, every developer’s MCP configuration is their own decision.
Where it works
Claude Code, the Claude desktop app, Cowork and the Agent SDK are all MCP clients. Because Claude Code shares configuration across its surfaces, a server configured once is available from the terminal, the IDE, the desktop app and the web.
Plan availability
Anything left blank is not documented publicly, not a claim that it is unavailable. Compare every feature at once in the availability matrix.
- FreeLimited
- ProIncluded
- MaxIncluded
- TeamIncluded
- EnterpriseIncluded
- APIIncluded
Comparisons
- MCP vs ConnectorsIf the person connecting it will never open a config file, you want a connector. If it is a developer wiring Claude Code or an SDK agent into an internal system, you want MCP. Many connectors are MCP underneath — the difference is who operates them, not what they are.
- Agent Skills vs MCPA Skill is knowledge — instructions and scripts Claude loads when the task calls for them. An MCP server is a connection — access to a tool or data source. If your problem is "Claude does not know our process", write a Skill. If it is "Claude cannot reach our system", stand up an MCP server.
Sources
- 01Introducing the Model Context ProtocolOfficialanthropic.com
- 02Connect Claude Code to tools via MCPOfficialcode.claude.com
- 03Control MCP server access for your organizationOfficialcode.claude.com
Page changelog
- VerifiedRe-checked against Claude Code's MCP and managed-MCP documentation.
- AddedMCP released as an open standard with reference servers and SDKs.