Tools & EcosystemHow It Works

MCP for operators: what it means and when you need it

In brief

MCP is the plumbing that lets Claude connect to anything. Here is what operators need to understand — and when it becomes relevant for your organisation.

5 min read·Model Context Protocol

Contents

♡Sign in to save

MCP — the Model Context Protocol — is an open standard that defines how Claude (and other AI models) connect to external tools and data sources. If you have used Connectors to link Claude to Notion or Google Drive, you have used MCP infrastructure, even if the word never appeared in the interface.

For most operators using Claude.ai, MCP is invisible — it is the plumbing behind features you use without thinking about them. For operators building custom integrations or using Claude via the API, understanding MCP becomes relevant.

What MCP actually does

Before MCP, connecting an AI model to an external tool required custom engineering for every integration — a different approach for every data source, every tool, every company. MCP standardises this: it defines a common format for how Claude requests data from external systems and how those systems respond.

The practical effect: any tool that builds an MCP server can be connected to Claude, without Anthropic needing to build a custom integration. This is why the ecosystem grew as fast as it did — tools connect themselves rather than waiting for Anthropic to connect them. The connectors directory now lists over 950 MCP servers.

The protocol itself has kept moving. The 2026-07-28 revision moved MCP to a stateless request/response core, which means servers can run on serverless and edge infrastructure rather than holding open a bidirectional session. If you are commissioning integration work, that is the revision to build against.

What this means for operators

If you use Claude.ai with Connectors: You are already using MCP. The Google Drive, Notion and Slack connectors all run on it. You do not need to understand MCP to use them.

What has changed and does deserve your attention: many connectors now write, not just read. The Microsoft 365 connector can draft and send email, manage calendar events, and create and update files. Airtable can create and update records. Treat authorising a connector as granting an access level, not as switching on a convenience — scope it to the workspaces and bases Claude actually needs.

If you want to connect Claude to an internal tool that doesn't have a built-in Connector: This is where MCP becomes operational for you. If your company uses a proprietary CRM, a custom knowledge base, or any internal system with an API, you can build an MCP server that lets Claude connect to it. This typically requires a developer — it is not a no-code task — but it is significantly less work than building a custom AI integration from scratch. Custom remote MCP connectors work on every plan, free through Enterprise.

If you are building AI-powered tools for your organisation: MCP is the standard your developers should use for any integration work. Building to MCP means your integrations work with the broader ecosystem, not just with Claude.

The ecosystem implication

Because MCP is an open standard, there is a growing library of pre-built MCP servers for common tools. Before having your developer build a custom integration, check whether an MCP server already exists for your tool. Many common development tools (GitHub, Linear, Jira, databases) have community or official MCP servers.

The connectors directory is the reviewed list, and anything in it can be added to Claude Code with claude mcp add as well. Check there before commissioning a build.

One governance note for anyone on Team or Enterprise: admins can control connector access through custom roles, down to which individual tools within a connector a given role can use. If you are the person answering "can we let the sales team connect Claude to the CRM," that is now a permissions question with a real answer rather than an all-or-nothing decision.

What operators don't need to worry about

If you are a non-technical operator using Claude.ai with standard Connectors, MCP is background infrastructure. The relevant question for you is: "Is the tool I want to connect available as a Connector in Claude.ai?" If yes, use it. If no, and it is a tool your whole organisation depends on, that is a conversation to have with your IT team about building an integration.

MCP is the reason that conversation is worth having — the integration path exists and is standardised. Connecting a proprietary internal tool to an AI model used to require significant custom engineering. With MCP, it requires a developer and a few days of work, and the result works with the wider ecosystem rather than only with Claude.

The honest summary

MCP is the infrastructure layer that makes Claude extensible beyond what Anthropic has built directly. For most operators, it is invisible. For organisations with proprietary tools they want Claude to access, it is the path from "Claude can't connect to our internal systems" to "Claude can connect to anything with an API." The technical barrier is real but much lower than it used to be.