Your apps, in any agent

Paste one line into your agent. It reads this page, sets up ClawLink and asks you to approve once in your browser.

Paste this into your agent

Start for free

1,500+ apps
Hosted OAuth
No API keys

Pick your agent

The same ClawLink account works in all of them. Connect an app once and every agent you pair can use it.

OpenClawInstall the plugin from ClawHub, then pair once from the chat.
HermesInstall the plugin, then approve one link in your browser.
Grok BotGrok Bot runs the CLI in its sandbox. The pairing sticks between sessions.
ClaudeOne connector on claude.ai and the desktop app, one command in Claude Code.
CodexCodex runs the CLI in its own terminal. You approve once.
MuseMuse installs ClawLink on its own cloud VM. You approve once.
Cursor, Windsurf and othersAny agent that can run a terminal command pairs with the ClawLink CLI.

This page is the setup guide. Your agent follows it.

Your agent finds its own section below, installs ClawLink and stops whenever you need to approve something.

claw-link.dev/agents

## For AI agents reading this page

You are reading https://claw-link.dev/agents. ClawLink connects you to 1,500+ apps (Gmail, Google Drive,
Notion, Slack, GitHub, and more) through hosted OAuth, plus public data (TikTok, Instagram, YouTube, Reddit, Amazon, crypto and stock prices, SEO) that needs no
sign-in at all. There are no API keys to collect.

First work out which agent you are, then follow only that section:

- OpenClaw: the "OpenClaw" section
- Hermes: the "Hermes" section
- Grok Bot: the "Grok Bot" section
- Claude: the "Claude" section
- Codex: the "Codex" section
- Muse: the "Muse" section
- Cursor, Windsurf, Claude Desktop, or any other agent that can run a shell command: the
"Any other agent" section

If you cannot tell which agent you are, ask the user which app they are chatting in.

### OpenClaw

Full page: https://claw-link.dev/openclaw

1. Install the plugin. In a terminal, run:

openclaw plugins install clawhub:clawlink-plugin

2. Pair this machine. Type this in the OpenClaw chat, not in a terminal:

/clawlink pair

It prints an approval link. Show the user the link and stop. Do not wait on it. The user
approves it in their browser and replies "approved".

3. Connect an app. When the user says approved, call the `clawlink_connect_app` tool with
the app they asked for (for example `gmail`). It returns a sign-in link. Show it and
stop again.

4. Find the tool for the job. `clawlink_list_integrations` lists the apps,
`clawlink_list_actions` lists what one can do, and `clawlink_execute` runs it.

The user signs in on ClawLink's hosted page, never inside this chat. Never enter
their credentials for them, and never approve access on their behalf.

Verify before you continue. Run:

clawlink_whoami

A pass names the account and its connected apps. If it fails, stop and tell the user what
failed. Do not carry on as if the tools were there.

Then tell the user which app is connected and what you can now do with it, and ask what they
want first. Repeat `clawlink_connect_app` for each new app; pairing runs once.

### Hermes

Full page: https://claw-link.dev/hermes

1. Install the plugin. Run:

hermes plugins install ClawLink-HQ/hermes-plugin --enable

2. Start pairing. Run:

hermes clawlink begin

It prints an approval link. Show the user the link and stop. Do not wait on the command.
The user approves it in their browser and replies "approved".

3. Finish pairing. When the user says approved, run:

hermes clawlink finish

4. Connect an app. Ask ClawLink to connect the app the user named (for example `gmail`).
It returns a sign-in link. Show it and stop again.

The user signs in on ClawLink's hosted page, never inside this chat. Never enter
their credentials for them, and never approve access on their behalf.

Verify before you continue. Run:

hermes clawlink test

A pass names the account and its connected apps. If it fails, stop and tell the user what
failed. Do not carry on as if the tools were there.

Then tell the user which app is connected and what you can now do with it, and ask what they
want first. Repeat `hermes clawlink` for each new app; pairing runs once.

### Grok Bot

Full page: https://claw-link.dev/grokbot

1. Pair this machine. Run:

npx -y @useclawlink/cli login

It prints an approval link. Show the user the link and stop. Do not wait on the command.
The user approves it in their browser and replies "approved".

2. Connect an app. When the user says approved, run:

npx -y @useclawlink/cli connect gmail

(replace gmail with the app they asked for). It prints a sign-in link. Show it and stop
again.

3. Find the tool for the job. Run `npx -y @useclawlink/cli apps <word>` to find an app,
`npx -y @useclawlink/cli actions <app>` to list what it can do, and
`npx -y @useclawlink/cli run <app> <action> --input '{...}'` to run one.

The user signs in on ClawLink's hosted page, never inside this chat. Never enter
their credentials for them, and never approve access on their behalf.

Verify before you continue. Run:

npx -y @useclawlink/cli whoami

A pass names the account and its connected apps. If it fails, stop and tell the user what
failed. Do not carry on as if the tools were there.

Then tell the user which app is connected and what you can now do with it, and ask what they
want first. Repeat `connect` for each new app; pairing runs once.

The pairing is stored at ~/.clawlink/credentials.json inside Grok Bot's sandbox and survives
between sessions.

### Claude

Full page: https://claw-link.dev/claude

**On claude.ai or the Claude desktop app** there is nothing to install. The user opens
Settings → Connectors → Add custom connector, names it ClawLink and enters the URL
https://claw-link.dev/api/mcp. Claude signs in to ClawLink itself over OAuth. Send them to
https://claw-link.dev/claude and let them click the button; do not ask them for an API key.

**In Claude Code**, do this instead:

1. Pair this machine. Run:

npx -y @useclawlink/cli login

It prints an approval link. Show the user the link and stop. Do not wait on the command.
The user approves it in their browser and replies "approved".

2. Add ClawLink to Claude Code. When the user says approved, run:

npx -y @useclawlink/cli install claude-code

It writes the MCP client entry and reads the credential login stored, so no API key is
pasted anywhere. By hand, the same thing is:

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

3. Connect an app. Run:

npx -y @useclawlink/cli connect gmail

(replace gmail with the app they asked for). It prints a sign-in link. Show it and stop
again.

4. Restart the session. The ClawLink tools appear on the next session, not the current one.
Tell the user to restart Claude Code, then use the tools directly.

The user signs in on ClawLink's hosted page, never inside this chat. Never enter
their credentials for them, and never approve access on their behalf.

Verify before you continue. Run:

npx -y @useclawlink/cli whoami

A pass names the account and its connected apps. If it fails, stop and tell the user what
failed. Do not carry on as if the tools were there.

Then tell the user which app is connected and what you can now do with it, and ask what they
want first. Repeat `connect` for each new app; pairing runs once.

### Codex

Full page: https://claw-link.dev/codex

1. Pair this machine. Run:

npx -y @useclawlink/cli login

It prints an approval link. Show the user the link and stop. Do not wait on the command.
The user approves it in their browser and replies "approved".

2. Connect an app. When the user says approved, run:

npx -y @useclawlink/cli connect gmail

(replace gmail with the app they asked for). It prints a sign-in link. Show it and stop
again.

The user signs in on ClawLink's hosted page, never inside this chat. Never enter
their credentials for them, and never approve access on their behalf.

Verify before you continue. Run:

npx -y @useclawlink/cli whoami

A pass names the account and its connected apps. If it fails, stop and tell the user what
failed. Do not carry on as if the tools were there.

Then tell the user which app is connected and what you can now do with it, and ask what they
want first. Repeat `connect` for each new app; pairing runs once.

### Muse

Full page: https://claw-link.dev/muse

You run in your own Muse Secure VM. The user already chose ClawLink for their apps. You are the
agent that will use these tools, so install and pair ClawLink in your own terminal, not on the
user's computer. The user's part is to approve links in a browser.

1. Start login in the background. Your shell stops a command after about 100 seconds, and login waits up to 15 minutes for the user's approval:

nohup npx -y @useclawlink/cli login > ~/clawlink-login.log 2>&1 &

2. After about 10 seconds, run `cat ~/clawlink-login.log`. Send the user the sign-in link and code, then wait until the user says "approved".

3. Check it: `npx -y @useclawlink/cli whoami` must name the user's account. If it fails, show the user the log. Do not act as if the tools are there.

4. Connect an app: `npx -y @useclawlink/cli connect gmail` (or the app the user asks for). Send the user the link and wait again. Public data apps (their names end in -data, for example `tiktok-data` or `crypto-market-data`) need no connect step: run them directly. Each call is paid from the user's ClawLink credit.

5. Use it:
- `npx -y @useclawlink/cli apps <word>` finds an app
- `npx -y @useclawlink/cli actions <app>` lists what it can do
- `npx -y @useclawlink/cli run <app> <action> --input '{...}'` runs one

If your network gate asks about claw-link.dev or registry.npmjs.org, ask the user to allow it. The login is saved in ~/.clawlink/ on your VM, so it stays between chats. Never type the user's passwords and never approve a link for the user.

### Any other agent

1. Pair this machine. Run:

npx -y @useclawlink/cli login

It prints an approval link. Show the user the link and stop. Do not wait on the command.
The user approves it in their browser and replies "approved".

2. Connect an app. When the user says approved, run:

npx -y @useclawlink/cli connect gmail

(replace gmail with the app they asked for). It prints a sign-in link. Show it and stop
again.

The user signs in on ClawLink's hosted page, never inside this chat. Never enter
their credentials for them, and never approve access on their behalf.

Verify before you continue. Run:

npx -y @useclawlink/cli whoami

A pass names the account and its connected apps. If it fails, stop and tell the user what
failed. Do not carry on as if the tools were there.

Then tell the user which app is connected and what you can now do with it, and ask what they
want first. Repeat `connect` for each new app; pairing runs once.

If you are Cursor, Windsurf or Claude Desktop, add ClawLink to your MCP config after pairing
so its tools show up natively:

npx -y @useclawlink/cli install <client>

where <client> is one of cursor, windsurf, claude-desktop, claude-code. Then ask the user to restart the
app. A client that adds remote MCP servers by URL and signs in with OAuth can instead add
https://claw-link.dev/api/mcp; no API key is needed.