--- id: connecting-claude-to-your-tools lang: en title: "Connecting Claude to your tools" seoTitle: "Claude connectors (MCP) explained" description: "Connect Claude to outside apps with connectors and MCP so it can act across your stack, from email to your CRM, and how to set them up safely." part: working-smart partTitle: "Working smart" partOrder: 4 order: 5 level: Intermediate durationMinutes: 6 outcome: "Connect Claude to the apps you use so it can act across them, and set it up safely." summary: "Connectors and MCP let Claude act across your other apps, from email to your CRM, so it does more than talk. Set them up narrowly and safely." draft: false updated: "2026-07-26" furtherReading: - title: "Use connectors to extend Claude's capabilities" url: "https://support.claude.com/en/articles/11176164-use-connectors-to-extend-claude-s-capabilities" note: "Anthropic's step-by-step: browse, connect, and manage connectors." - title: "Connectors Directory" url: "https://claude.ai/connectors" note: "The full list of apps you can connect, each with its read/write capabilities." - title: "Get started with custom connectors using remote MCP" url: "https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp" note: "How to connect a service that isn't in the directory." - title: "Model Context Protocol" url: "https://modelcontextprotocol.io" note: "The open standard connectors are built on, if you want the deeper story." keywords: - Claude connectors - what is MCP - connect Claude to apps - Claude integrations - Model Context Protocol - AI tool integration quiz: intro: "Five quick questions. Ace them all and you've finished Part 4." passScore: 5 questions: - q: "What does a connector let Claude do?" options: - "Make Claude run faster" - "Reach into an outside app to read your data and take actions there, like searching Drive or sending a Slack message" - "Change which model you're using" - "Store your password" answer: 1 explain: "A connector is a bridge to an app you already use. Claude can pull your data in (read) and do things back (act), so you stop being the courier between Claude and your tools." - q: "A connector wants to reach a file you personally can't open in the source app. What happens?" options: - "Claude sees it anyway" - "It can't reach it either: Claude inherits your permissions, so it only ever sees what you can see" - "Claude asks the file's owner" - "The connector deletes the file" answer: 1 explain: "This is the golden rule. A connector never grants more access than you already have. If you can't open it, neither can Claude through the connector." - q: "Why treat 'read' actions and 'act' actions differently?" options: - "Reading costs more money" - "Reading just looks something up, while acting (send, create, delete) changes the real world, so it deserves more caution" - "There's no difference" - "Acting is always blocked" answer: 1 explain: "Searching or summarising is low-risk. Sending a message, creating a record, or deleting something is not. Enable write-capable connectors deliberately, and on Team or Enterprise, admins can lock one to read-only." - q: "What is MCP, in plain terms?" options: - "A paid Claude plan" - "The open standard connectors are built on: a universal plug, so many different apps connect the same way" - "A type of file" - "A password manager" answer: 1 explain: "MCP (Model Context Protocol) is the shared standard behind connectors. Because it's open, hundreds of apps work the same way, and any service that speaks MCP, even one your company builds, can connect." - q: "You've got a dozen connectors switched on and chats feel sluggish. What's the fix from lesson 4.1?" options: - "Buy a bigger screen" - "Connectors are token-heavy, so enable only the ones a chat needs, or switch tool access to 'On demand'" - "Delete your account" - "Nothing can be done" answer: 1 explain: "Every active connector takes up room in the window. Keep just the ones you need on for a given chat, and with ten or more, 'On demand' loading gives your conversation more space." pass: badge: "Part 4 complete" title: "You work smart now." body: "Full marks, and that's all of Part 4. You can manage context, package Skills, run Projects, hand tasks to Cowork, and connect Claude to your apps. Next comes Part 5: the advanced, specialised stuff, from Claude Code to automation and full playbooks." retry: "Check the marked answers and give it another go. You've got this." --- > **Level:** Intermediate ยท **Reading time:** ~6 min > > **What you'll be able to do:** connect Claude to the apps you use so it can act across them, and set it up safely. *(Last checked: July 2026.)* Up to now, Claude works with what you hand it: files you upload, a folder you connect. Useful, but you are still the courier, fetching things from your email, your Drive, your project tracker, and carrying the results back. A **connector** cuts out the courier. It lets Claude **reach into the apps you already live in** and work there directly. ## What a connector is A connector is a **bridge between Claude and an outside app.** Once it is in place, Claude can do two kinds of thing: - **Read:** pull your data in. "Find the Q3 budget in my Drive." "What did the client say in Slack yesterday?" - **Act:** do things back. "Create a Linear issue for that bug." "Draft and send this Slack message." "Add the meeting to my calendar." Connect Claude to Google Drive and it searches your files. Connect it to Slack and it posts messages. Connect it to your issue tracker and it opens tickets. The work stops bouncing through you. ## MCP, in one breath You will see the letters **MCP** attached to all this. It stands for **Model Context Protocol**, and it is the open standard connectors are built on. Skip the acronym and picture a **universal plug.** Because every connector uses the same standard, hundreds of apps connect the same way, and *any* service that speaks MCP can join, including a tool your own company builds. That is the whole point: one plug, many apps.
Claude at the centre, your apps around it Claude sits in the middle. Connectors link it out to apps like Google Drive, Gmail, Slack, Calendar, and an issue tracker. Each link lets Claude read data from the app and take actions in it. One hub, many apps Claude Drive Gmail Slack Calendar Issue tracker Each link lets Claude read from the app and act in it. Over 375 apps are in the directory.
## The golden rule: it only sees what you see Before anything else, the safety cornerstone: **Claude inherits your permissions.** A connector can never reach a file, a channel, or a record that you could not open yourself in the source app. If a Drive folder is off-limits to you, it is off-limits to Claude too. Connecting an app does not hand Claude the keys to the whole company, only your own set. ## Setting one up It takes a couple of minutes: 1. **Browse the Connectors Directory** (at claude.ai/connectors, over 375 apps). Each has a page showing what it does and its read and write capabilities. 2. **Pick one and read its capabilities**, so you know what you are granting. 3. **Click Connect and sign in.** You authenticate with the app and approve the access it asks for. Standard, secure sign-in, the same as logging into the app anywhere. 4. **Switch it on in a chat** with the **+** button. From then on, Claude can even bring a connected app in on its own when your request calls for it, without you naming it. Not in the directory? You can add a **custom connector** by entering its MCP address, handy for a service your company runs. (Free accounts get one custom connector; paid plans, more.) ## Connections worth setting up first The directory is huge, so start where your day actually is. A handful that pay off fast for most business work: - **Email (Gmail or Outlook):** *"Summarise my unread email from this morning and draft replies to the three that need one."* Claude reads the inbox and hands you drafts to check and send. - **Calendar (Google or Outlook):** *"Look at my week and block two focused hours for the board deck."* It sees your schedule and proposes the slot. - **Files (Google Drive, Microsoft 365, Box, Dropbox):** *"Find the latest Acme contract and pull out the renewal date and payment terms."* It searches your storage instead of you hunting for the file. - **Chat (Slack or Microsoft Teams):** *"Catch me up on the #project-atlas channel since Friday, then post a three-line decision summary."* Reading and posting, in one go. - **CRM (Salesforce, HubSpot):** *"Before my 2pm call, pull this account's open deals and the last three touchpoints."* Your sales prep, done while you grab coffee. - **Project tracker (Jira, Linear, Asana):** *"Turn these meeting notes into tickets, one per action item, with owners and labels."* Notes in, a tidy backlog out. - **Docs and wikis (Notion, Confluence):** *"Draft the kickoff page from these notes and file it under the team space."* It writes and files in the same step. Notice the pattern: the real wins pair a **read** ("catch me up", "find", "pull") with an **act** ("draft", "post", "create"). That is where a connector saves you the most time, and it is also where the caution in the next section matters most. (Exactly which apps and capabilities are available shifts often, so check the directory for the current list.) ## Read versus act: know the line The two verbs from earlier are not equal in risk, and this is the judgement call worth making every time.
Read is low-risk, act changes the world Two columns. Read actions like search, summarise, and look up are low-risk. Act actions like send, create, edit, and delete change the real world and deserve more caution. Two verbs, two risk levels Read Search a Drive Summarise a thread Look up a record Low-risk. Just looks. Act Send a message Create or edit a record Delete something Higher-risk. Changes things.
Reading looks something up and nothing more. **Acting changes the real world:** a message goes out, a ticket appears, a file is gone. So enable write-capable connectors *deliberately*, and lean on the oversight thinking from the Cowork lesson: match how closely you watch to what is at stake. On **Team and Enterprise** plans, an admin can lock a connector to **read-only** for everyone, a clean way to get the lookups without the risk. ## Keep it lean and safe Five habits, most of them familiar by now: - **Only turn on what a chat needs.** Connectors are **token-heavy** (remember lesson 4.1): every active one eats window space. With ten or more, switch tool access to **On demand** so your conversation keeps its room. - **Only connect apps you trust.** Directory connectors are vetted; a **custom** connector is not, so add one only from a source you trust. - **Mind where your data goes.** A connected service handles your data on **its own systems, under its own terms**, which may sit outside your country. The privacy lens from lesson 1.4 still applies: connecting is sharing. - **Match oversight to the stakes**, exactly as in Cowork. Stay close to anything that sends, spends, or deletes. - **Disconnect what you stop using.** An unused connector is standing access you have forgotten about. Tidy it away. ## That's Part 4 Step back and look at what you can do now. You can keep a long chat sharp by **managing context** (4.1). You can package a repeatable job as a **Skill** (4.2). You can give an ongoing piece of work a home in a **Project** (4.3). You can hand a whole task to **Cowork** and let it work your files (4.4). And now you can **connect Claude to the apps** you already use, so it acts across your stack (4.5). That is the "working smart" kit, and together it turns Claude from a clever chat into a genuine part of how you work. **Part 5** goes further and more specialised: Claude Code for the technical, Claude for design, scheduling and automation, and full end-to-end playbooks that chain everything you have learned. ## Try it Name the **two apps you touch most** in a working day. For each, write down one **read** job Claude could do ("find last week's numbers") and one **act** job ("post the summary to the team channel"). Now the useful part: decide which of those you would keep **read-only** for now, and which you would trust Claude to act in. That single call, where you draw your own line between looking and doing, is the heart of connecting tools safely. *(General exercise. Role and industry versions come once accounts are in.)* ## Key terms [Connector](/glossary#connector), [MCP](/glossary#mcp), [Context window](/glossary#context-window), [Project](/glossary#project)