What Are Claude Connectors? A Plain Explanation

Blog · MCP & agents

What Are Claude Connectors? A Plain Explanation

Claude connectors are one-click integrations in Claude’s settings, each an MCP server under the hood. What they do, how to add one, and how safe they are.

· 7 min read · by the LedgerMCP team

Claude connectors are one-click integrations you add from Claude’s Settings, under Connectors. Each one links Claude to an outside app, your files, your email, your books, so Claude can read from it and act in it during a normal chat. Under the hood, every connector is an MCP server. That last part is the detail most explanations skip, and it is the one that makes the whole system make sense. Here is what connectors do, how to add one, and what to check before you grant one access.

What do Claude connectors actually do?

Out of the box, Claude only knows what you paste into the chat. A connector removes that wall for one specific app. Connect Google Drive and Claude can search your documents. Connect a payments tool and it can look up charges. Connect a bookkeeping ledger and it can post entries and run reports. Anthropic launched the connectors directory in July 2025, and as of August 2026 it lists well over fifty integrations, spanning files, email, design, payments, project tracking, and accounting. The point is not any single integration. It is that Claude stops being a text box you copy things into and starts being an assistant that works inside the tools where your actual data lives.

Are connectors the same thing as MCP servers?

Yes, and the distinction is packaging, not technology. MCP (Model Context Protocol) is the open standard for exposing an app’s actions as named tools an assistant can call. An MCP server is any piece of software that publishes those tools. A connector is an MCP server that has been listed in Claude’s directory and wrapped in a Connect button, so instead of pasting a server URL or editing a config file, you click once and sign in. If you want the deeper technical picture, start with what an MCP server is. The practical takeaway: when you compare "connectors vs MCP," you are not choosing between two systems. Connectors are MCP with the setup friction removed.

How do you add a connector?

The whole flow takes a minute or two:

  1. In Claude, open Settings, then Connectors, and browse the directory.
  2. Find the app you want, for example the LedgerMCP bookkeeping connector, and click Connect.
  3. Complete the sign-in for that service and grant access. For LedgerMCP it is your email plus a 6-digit code.
  4. Start a chat and ask Claude to do something in the connected app. Claude calls the connector’s tools and shows you the result.

No URL pasting, no JSON config. That used to be the setup story for MCP, and for developer tools like Claude Code or Cursor it still is (those connect with an API key or a server URL instead). For the assistant most people use, the directory replaced all of it.

Remote connectors vs local ones

Most directory connectors are remote: the MCP server is hosted by the company behind the app, and Claude talks to it over the network. Remote connectors work from claude.ai in a browser and on mobile, with nothing installed on your machine. A smaller set are local, desktop-app connectors that run on your computer and let Claude work with software installed there. For everyday services, remote is what you want: it is the version where "works everywhere, nothing to maintain" is literally true. Local matters mainly for developers and for apps whose data never leaves the machine.

Are connectors safe? What can they actually touch?

A connector can only touch what you authorize. You sign in to the service yourself, the connector gets the access that sign-in grants, and Claude asks before taking consequential actions rather than acting silently. Beyond that baseline, safety varies by connector, so it is worth checking what guardrails the service itself enforces. Good signs: read-only modes for when you only want questions answered, previews before writes, and actions that can be reversed. LedgerMCP, for instance, offers read-only API keys, dry-run previews on writes, and makes every write reversible, with the accounting invariants enforced in the database rather than trusted to the model. The other half of safety is restraint: connect the two or three apps you will actually use, not everything in the directory. Our post on how AI agents find MCP servers covers how discovery and trust work in the broader ecosystem.

A worked example: the LedgerMCP bookkeeping connector

Here is what a connector looks like end to end. LedgerMCP is listed in Claude’s connector directory: a free double-entry bookkeeping ledger with 65 MCP tools, 25 read and 39 write. You connect it from Settings, sign in with your email and a 6-digit code, and an account is created automatically. From then on, in any chat, Claude can import a bank CSV, categorize transactions, split and transfer, reconcile an account, and run a P&L or balance sheet, each step a real posted entry in a real ledger. LedgerMCP has no built-in AI of its own; Claude supplies the judgment, the connector supplies the accounting engine and the guardrails. The software is free with unlimited books; the only paid part is optional Plaid bank feeds, from $9 a month. The full setup and what to try first is on the Claude connector page, and the walkthrough post on connecting Claude to an accounting MCP goes step by step.

Quick answers

What are Claude connectors?

Connectors are one-click integrations listed in Claude’s Settings under Connectors. Each one links Claude to an outside app, like your files, your calendar, or your bookkeeping ledger, so Claude can read from it and, where allowed, act in it during a chat.

Are Claude connectors the same as MCP servers?

Under the hood, yes. Every connector is an MCP server, the open standard for giving assistants tools. The directory just packages the server behind a Connect button so you sign in instead of pasting a URL or editing a config file.

Are Claude connectors safe?

The model is permission-based: you sign in to each service yourself, Claude only gets the access that sign-in grants, and it asks before taking consequential actions. Good connectors add their own guardrails on top, like read-only keys and reversible writes. Only connect what you will actually use.

Do Claude connectors cost anything?

Adding a connector is free. What can cost money is the service behind it: the QuickBooks connector expects a paid QuickBooks subscription, for example, while LedgerMCP’s software is free and only its optional bank feeds are paid.

Put this into practice

Free books in one minute: connect Claude or ChatGPT and let it do the work you just read about.