MCP & connectors

On its own, Claude only knows what's in the conversation and its training. Connectors, built on the Model Context Protocol (MCP), let Claude reach into your real tools and data that are live, current, and specific to you.

What is MCP?

The Model Context Protocol (MCP) is an open standard for connecting AI assistants to external tools and data sources. Think of it as a universal adapter: instead of every app needing a custom integration, MCP gives Claude a standard way to talk to many systems.

MCP has become the de-facto standard for agent connectivity, with broad industry adoption.

What it unlocks

With connectors, Claude can do things like:

• Search and read from your knowledge base or documents.
• Pull live data from a database or API.
• Look up information in tools like issue trackers, CRMs, or docs.
• Take actions in connected systems (e.g. create a ticket), where permitted.

The result: answers grounded in your current reality, not just the model's general knowledge.

Connectors in everyday use

In claude.ai (on supported plans), you can enable connectors to your tools. Once connected, you just ask naturally:

Check our docs and summarize our current refund policy.

What are the open high-priority issues assigned to my team this week?

Claude uses the connector to fetch the real data, then reasons over it.

MCP for builders

Developers can connect Claude to custom MCP servers, wrapping internal APIs, databases, or services, so Claude-powered apps and agents operate on live systems. There's even a "MCP Builder" Skill that encodes the patterns for building these integrations correctly.

Skills vs. MCP: a quick distinction

People mix these up:

• Skills = know-how. They teach Claude how to do a task (procedures, standards).
• MCP = connectivity. It gives Claude access to tools and data.

They're great together: a Skill might define how to write a competitive analysis, while MCP supplies the live web, CRM, and document data it needs to fill it in.

Safety & permissions

Connectors can read, and sometimes act

Only connect trusted tools, review what each connector is allowed to access, and be cautious with connectors that can take actions. Grant the least access needed. See Privacy & data.

Next: the developer-focused Claude Code.