Claude Connectors: How to Add Them, Plans and Limits

In brief

Claude connectors are integrations that let Claude read data and take actions inside outside apps such as Google Drive, Gmail and Slack, built on the Model Context Protocol. Anthropic's directory listed 1,127 of them on 15 November 2026. You add one under Customize, then Connectors, and Free accounts can add one custom connector.

What are Claude connectors?

Claude connectors are links between Claude and the apps you already use. Anthropic's help center defines them in one line: "Connectors let Claude access your apps and services, retrieve your data, and take actions within connected services." Its own examples are creating issues in Linear, sending messages in Slack and searching your files in Google Drive. They run on MCP, the open standard explained in what is MCP, and they work in Claude on the web, Claude Desktop, Claude Code and the API.

A terracotta power strip on a charcoal plinth with five different plugs, cables running to small clay objects shaped like a folder, an envelope, a calendar and a speech bubble, beside a glowing glass panel carrying the Claude mark.

The directory is the shop window. Browse it first. Anthropic announced it on 14 July 2025 with Notion, Canva and Stripe as remote connectors and Figma, Socket and Prisma as local desktop ones. On 15 November 2026 the page at claude.com/connectors, which now redirects to Anthropic's Claude Marketplace and also lists plugins, showed 1,127 connectors across 17 categories, from Productivity and Developer tools to Legal, Healthcare and Travel. You can filter by type: Web, Desktop extension or Interactive. Interactive connectors draw live dashboards, task boards or design tools inside the chat instead of returning text.

Claude connectors went from one launch post to 1,127 listings
20252026ProtocolAnthropic publishes MCPMCP moves to the Linux FoundationClaudeConnectors Directory launchesDirectory lists 1,127 connectors

One rule governs all of it. Claude acts with your permissions and nobody else's. The help center says "Claude inherits each person's permissions from the connected service," so a connector can't open a record you can't open yourself. The one exception is a custom connector that uses a shared credential such as an API key, and that connector reaches whatever the key reaches. On a Team or Enterprise plan that holds for every member who uses it. Keep keys narrow.

How do you add and use a connector in Claude?

You add a connector from the Connectors Directory or from Claude's settings, then switch it on per chat. The steps below come from Anthropic's help center as read on 15 November 2026.

  1. Open Customize, then Connectors, and click + next to Connectors. In a chat you can get there faster: click + in the lower left corner (or type /), hover over Connectors, then pick Manage connectors.
  2. Find the service in the directory and choose Connect. Each listing has its own page with use cases, read and write abilities and availability.
  3. Sign in to the service and read what access it asks for. You authenticate as yourself, so your account's limits apply.
  4. Back in a chat, click + (or type /), hover over Connectors, and toggle on the services you want for that conversation.
  5. Ask in plain words, such as "Create a new issue for the login bug" after connecting Linear. Once an app is connected, Claude can also bring it into a chat on its own when the request fits.
  6. Approve what Claude asks to do. Each connector lists its tools, grouped by type, and the help center says to review approval requests and use Allow always only for servers and tools you trust.

Connectors you set up on Claude or Claude Desktop show up on Claude for iOS and Android the next time you log in, so you never repeat the setup on your phone. Installing connectors on mobile itself is in beta, and Anthropic calls Claude Desktop and the web the main route for custom ones. If you connect 10 or more, Anthropic suggests changing Tool access from Auto to On demand, which gives each conversation more room. You find it in the same + menu, under Connectors, then Tool access. To remove a connector, go back to Customize, then Connectors, and choose Remove.

Which Claude plans get connectors, and what does Free get?

Free accounts get one custom connector, and every paid plan gets custom connectors without that cap. That is the plan rule Anthropic states in its custom connector guide, as of 15 November 2026. Anthropic's help center also says web connectors from the directory are available to all users, and desktop extensions to all users on Claude Desktop. Each listing carries its own availability on its connector page, so check the listing before you count on one.

PlanCustom connectors (remote MCP)Who switches it on
Free1You
Pro, Max, UltraYesYou
TeamYesAn Owner or Primary Owner adds it, then each member signs in
EnterpriseYesAn Owner, Primary Owner or a custom role with Manage access to Libraries adds it, then each member signs in

The history matters if you read older guides. In the 14 July 2025 launch post Anthropic wrote: "Connectors to remote apps and services are available to paid plan users only." The help center now lists custom remote connectors on Free as well, with the one-connector limit. A page that still says remote connectors need a paid plan is repeating the launch wording.

On Team and Enterprise, the admin decides before anyone connects. Owners can restrict actions org-wide, for example allowing a connector to read from a service while blocking it from writing back, and individual users can't override that restriction once an owner has set it for the whole organization. Team plans have connector requests on by default so members can ask an owner to add one. On Enterprise that setting starts off.

How do custom connectors, desktop extensions and the API differ?

A custom connector points Claude at any remote MCP server, a desktop extension installs a local one, and the API takes a server in the request. All three do the same job for different readers. Only the first is an everyday feature for chat users, and the other two mostly matter if you build or host tools yourself.

To add a custom connector on Pro or Max, go to Customize, then Connectors, click + Add, then Add custom connector. (The general connectors guide words the same path as Add, hover Custom, then Web.) Enter a name and the server URL, check the detected authentication settings, and choose Sign in now, Sign in when needed or No sign in. Fixed credentials such as API keys go under Request headers and are stored encrypted. On Team and Enterprise an owner does the same under Organization settings, then Connectors, and members then click Connect.

The network rule trips people up. Custom connectors are reached from Anthropic's cloud, even if you use Claude Desktop on your own computer. The server must be reachable over the public internet on HTTPS port 443, and URLs with other ports aren't supported. A server on a private network can work if you allowlist Anthropic's IP addresses in your firewall. Local servers set up through claude_desktop_config.json are a separate mechanism that isn't available in claude.ai.

RouteWhat it isWhere you set it upMain limit
Directory connectorPre-built, reviewed by AnthropicCustomize > ConnectorsAvailability varies by listing
Custom connectorYour own remote MCP server URLCustomize > Connectors > Add custom connectorPublic HTTPS on port 443
Desktop extensionLocal MCP server in a one-click .mcpb packageSettings > Desktop app > Manage next to ExtensionsClaude Desktop only
API MCP connectorRemote servers called from the Messages APImcp_servers and mcp_toolset in the requestBeta, tool calls only, public HTTP

Desktop extensions use the .mcpb file format and run Node.js, Python or binary servers. Claude Desktop ships a built-in Node.js environment. The API version is in beta under the header mcp-client-2026-09-15, with mcp-client-2025-11-20 still listed. It supports only tool calls from the MCP spec, and the server must be public over Streamable HTTP or SSE, so local STDIO servers can't connect directly. If you code with Claude Code, Claude Code MCP covers that route.

What can go wrong with Claude connectors?

The biggest risk is prompt injection, and Anthropic says so in its own guide. Anthropic hasn't verified custom connectors, and a malicious server can hide instructions in the text it returns. The help center tells you to connect only to trusted servers, limit OAuth scopes where possible and review tool inputs and outputs. What is prompt injection explains why a model that reads untrusted text and can act is the whole problem.

Research raises the stakes. During Claude's advanced research, connector tools run automatically with no approval step, and the help center advises disabling any tool that writes. Many requests to your connectors can also have an impact on the service on the other end. Plan for that. Advanced research can't call tools from local MCP servers at all.

Three plain failures are worth knowing. First, a custom connector that times out is nearly always a firewall or a private network, because the connection starts in Anthropic's cloud. A corporate email on a domain an Enterprise organization has verified can trigger the message that says the identity belongs to an Enterprise that manages access through its own Claude account. And connected services process your data under their own terms, possibly outside the United States, which the US-only inference setting on Enterprise does not change.

Connectors don't make Claude smarter. They give it your data and your buttons. For the wider picture of models that act through tools, see what is an AI agent.

How do Claude connectors compare with ChatGPT and Gemini?

All three assistants now offer the same idea under different names, and only Claude's help page states a free-plan count for custom connections. Claude calls them connectors, ChatGPT calls them apps inside plugins, and Gemini calls them Connected Apps. The table is compiled from each vendor's own help page on 15 November 2026.

ClaudeChatGPTGemini
NameConnectorsApps and pluginsConnected Apps
Where to addCustomize > ConnectorsSettings > PluginsCustomize > Connectors in the sidebar
Use in a chat+ or /, then Connectors@ or +@ in the prompt box
Plan rule on the help pageCustom connectors on Free (1), Pro, Max, Ultra, Team, EnterpriseNo per-plan table: depends on app, plan, region, workspace, role, model and interfaceNo plan named: varies by location, language, device and app
Your own MCP serverYes, custom connectorNot covered on the apps pageYes, by linking an MCP server
Permission controlPer-tool approval, org-wide limits on Team and EnterpriseAlways ask, Allow read actions, Allow low-risk actions, Allow all actionsNot stated on the page

Choose by app, not by brand. Pick Claude if the app you need is in its directory or your own server speaks MCP, because the help center spells out the setup and the network rules. Pick ChatGPT if you want a fixed ladder of confirmation levels, and note that OpenAI itself says "Allow all actions carries elevated risk because supported actions may run without another confirmation." Pick Gemini if your life is already in Gmail, Calendar and Home, which it lists among its Connected Apps, and remember that Gemini needs Keep Activity switched on for some of them.

Whizi, which publishes this page, can't give you connectors: it has no app connectors and no MCP support, and web search, on by default in each chat, is the one outside source it reads. What it does cover is the chat half of working with several assistants. Pro carries Claude, GPT and Gemini in one app, with a first week of $1.99, and the free plan gives 10 messages a day on Whizi AI, Auto and Claude Haiku 5.5 with no card. Keep Claude's own app for Drive, Gmail and Slack, and use Whizi when you want to put the same question to a second model. ChatGPT and Claude in one app shows what carries over when you switch.

Workflow checklist
  • Add connectors from Customize, then Connectors, and switch each one on per chat through the + menu
  • Remember Claude uses your permissions: a connector can't open anything you can't open yourself
  • On Free, plan around one custom connector; on Team and Enterprise, ask an owner to add it first
  • Host a custom connector on the public internet at HTTPS port 443, or allowlist Anthropic's IP ranges
  • Connect only servers you trust, and use Allow always sparingly
  • Turn off write tools before you run advanced research, because connector tools run without approval there
  • Switch Tool access to On demand once you pass 10 connectors
Common questions

Frequently asked questions

Are Claude connectors free?

Partly. Anthropic's custom connector guide lists Free alongside Pro, Max, Ultra, Team and Enterprise and caps Free accounts at one custom connector. The July 2025 launch post said remote connectors were for paid plans only, and each directory listing states its own availability, so check the listing before you rely on it.

What is the difference between Claude connectors and MCP?

MCP is the open protocol and a connector is Claude's feature built on it. A connector is either a pre-built MCP server from the directory or a custom one you add by URL. The help center says connectors work in Claude, Claude Desktop, Claude Code and the API through the MCP Connector.

Do Claude connectors work on mobile?

Yes, once you set them up elsewhere. Anthropic says a service you connect on Claude or Claude Desktop becomes available the next time you log in on Claude for iOS or Android. Custom connectors still run from Anthropic's cloud, so the server must be public.

How many connectors does Claude have?

Anthropic's directory showed 1,127 connectors in 17 categories on 15 November 2026, split into Web, Desktop extension and Interactive types. The count moves as partners submit new ones, and you can also add any remote MCP server yourself as a custom connector.

Are Claude connectors safe?

They are as safe as the app and the server behind them. Claude can only reach what your account can reach, and Anthropic reviews directory listings. Anthropic hasn't verified custom connectors, warns that malicious servers can hide prompt-injection instructions, and says tool calls in advanced research run with no approval.