Connect apps to Claude Code

Claude Code reads tools from MCP servers you point it at. The default path is to build one per app: register an OAuth app in the provider's developer console, create a credentials file, run an MCP server locally, and declare it in Claude Code's config. Do that for Gmail, Slack, and Notion and you are maintaining three servers with three sets of tokens.

The alternative is a CLI that manages MCP registrations for you. The ClawLink CLI collapses the whole thing into a browser login and one install command:

npx -y @useclawlink/cli login
npx -y @useclawlink/cli install claude-code

login opens a ClawLink approval page; approve it and the CLI stores a credential in ~/.clawlink/credentials.json (the server keeps only a SHA-256 hash). install claude-code runs claude mcp add for you, pointing at a hosted MCP server. No API key appears in any config file — no MCP JSON to hand-edit, no server to run.

Then connect the apps you use:

npx -y @useclawlink/cli connect gmail    # opens Google's consent screen
npx -y @useclawlink/cli connect notion   # opens Notion's consent screen
npx -y @useclawlink/cli apps             # list the 230 available apps

The app's OAuth token never lands on your machine. ClawLink stores it server side, refreshes it when it expires, and exposes the app's actions as tools over MCP. You keep using Claude Code exactly as before — the connected apps just show up as tools in chat.

What Claude Code can do with a connected app

The tools come from the app's own API, exposed by name. A few examples:

• Gmail: gmail_send_email, gmail_search_messages, gmail_get_attachment — read, search, draft, send, and label mail.
• Notion: notion_create_page, notion_query_database, notion_search — read and write pages and databases you can access.
• Slack: slack_post_message, slack_search_messages, slack_list_users — read and send in the workspaces you are a member of.
• GitHub: github_create_pull_request, github_list_issues, github_create_issue — file PRs and triage issues from chat.

The write tools are the ones to be deliberate with: sending an email, filing a PR, or creating a page is a real action. Ask Claude Code to confirm what it is about to do before it does it, the same way you would with any terminal tool.

If you prefer to wire it yourself

install claude-code is a convenience. Claude Code can also talk to the ClawLink MCP server directly over HTTP, no shim in the path. Create an API key under Settings → Devices (cllk_live_...) and run:

claude mcp add --transport http clawlink https://claw-link.dev/api/mcp --header "x-clawlink-api-key: cllk_live_your_key_here"

For a stdio client, the equivalent is:

claude mcp add --transport stdio clawlink -- npx -y @useclawlink/mcp

Either way, the same hosted connection and the same token refresh apply.

Does this need a Google Cloud project or an API key?

No. The whole point of the hosted connection is that ClawLink owns the OAuth client, the scopes, and the refresh cycle. You approve the provider's consent screen once; the token lives server side and rotates without you noticing. There is no .env file, no credentials.json, and no key to paste into Claude Code's config.

This is different from the common "paste an API key" pattern: some apps do ship API keys (a provider like Postmark or SerpAPI has one), and in those cases you paste it once into the connection settings instead of a consent screen. Either way the key ends up server side, not in your MCP config.

Using OpenClaw or Hermes instead?

The same ClawLink account works across agents. For OpenClaw, install the OpenClaw plugin and pair with hermes clawlink begin / finish on the Hermes side; every connected app carries over. The connect apps to any AI agent page covers the full catalog.