# Install a WireCat CLI and MCP server for your AI agent (/en/docs/installation)

Connect the information you want your AI agent to use: messages, notes, email, documents or
meetings. Choose one tool below, install it where your agent can run it, and check a small result.
You do not need a messenger account to start with local files or a downloaded meeting transcript.

Already installed and connected? Go to [connect your agent](/llms.mdx/docs/agents/content.md), then
[try a first task](/llms.mdx/docs/content.md#try-a-request). An empty search usually needs a data-coverage check,
not another installation.

![Claude Code in a terminal installs a WireCat tool with npm; WireCat Desktop shows the Tools screen](/screenshots/install/install-and-connect-light-en.webp)

Rather not use a terminal? [WireCat Desktop](/en/download) is an app for macOS, Windows and Linux.
It installs and updates these tools, sets up Node.js when it is missing, gives your agents the tools'
skills and opens each tool's sign-in. [What the app does](/llms.mdx/docs/desktop/content.md).

**On this page:**

1. [Choose a tool](#choose-a-tool)
2. [Set up with your agent](#set-up-with-your-agent)
3. [Set up by hand](#set-up-by-hand)
4. [Node.js, npm and finding the command](#nodejs-npm-and-finding-the-command)
5. [Check the first useful result](#check-the-first-useful-result)
6. [Data, updates and background work](#data-updates-and-background-work)

## Choose a tool

| You want to use | Installation and connection |
|---|---|
| Notes, local files or email | [Install Memo](/llms.mdx/docs/memo/installation/content.md); a folder is enough to begin: see [Markdown and Obsidian](/llms.mdx/docs/markdown/content.md) |
| Telegram chats | [Install Telegram](/llms.mdx/docs/tg/installation/content.md); for a bot, use [bot setup](/llms.mdx/docs/tg/bot/content.md) |
| WhatsApp chats | [Install WhatsApp](/llms.mdx/docs/wa/installation/content.md); business numbers use [Cloud API bot setup](/llms.mdx/docs/wa/bot/content.md) |
| MAX chats | [Install MAX](/llms.mdx/docs/max/installation/content.md); for a bot, use [bot setup](/llms.mdx/docs/max/bot/content.md) |
| Google Drive documents | [Install Drive](/llms.mdx/docs/drive/installation/content.md) from npm, connect Google and choose a folder |
| Notion pages and databases | [Install Notion](/llms.mdx/docs/notion/installation/content.md) from npm, then [sign in with a Notion token and copy your pages](/llms.mdx/docs/notion/content.md#connect-and-copy-your-pages) |
| Yandex Disk files | [Install Yandex Disk](/llms.mdx/docs/ydisk/installation/content.md) from npm, then sign in with a Yandex token and [copy a folder](/llms.mdx/docs/ydisk/usage/content.md#sync) |
| Google Calendar meetings | [Install Google Calendar](/llms.mdx/docs/gcal/installation/content.md) from npm and sign in to Google through WireCat's Google app, or your own |
| Email over IMAP | [Install Mail](/llms.mdx/docs/mail/installation/content.md) from npm, then make an app password for your mailbox |
| Zoom meetings or transcripts | [Install Zoom](/llms.mdx/docs/zm/installation/content.md); connect an existing/new app or import VTT files without one |
| Google Meet calls and transcripts | [Install Google Calendar](/llms.mdx/docs/gcal/installation/content.md), then sign in with `--grant meetings`: see [Google Meet](/llms.mdx/docs/google-meet/content.md) |
| A Bitrix24 CRM | [Install Bitrix24](/llms.mdx/docs/b24/installation/content.md) from npm, then make an incoming webhook or a local application on your portal |
| An amoCRM or Kommo CRM | [Install amoCRM](/llms.mdx/docs/amo/installation/content.md) from npm, then make a private integration in your account and sign in with its token |
| Calls of an Asterisk or FreePBX phone system | [Install and connect Asterisk](/llms.mdx/docs/pbx/installation/content.md): install it from npm on your own computer, then connect the call records and recordings |

Each guide supplies its exact commands, account steps, supported platforms and repairs. A Zoom
app uses a Public Client ID; Google Drive's optional custom app can use a client JSON file; Google Calendar
signs in through WireCat's Google app, or a client JSON file of your own Google app. Follow
the chosen guide rather than looking for one credentials file shared by every tool.

## Set up with your agent

Use an AI agent (An AI agent that can run commands on your computer, for example Claude Code, Codex, Cursor or Gemini CLI. A chat in the browser without terminal access cannot install the command-line tool for you.) with terminal access on this computer, such as Claude Code, Codex, Cursor or Gemini
CLI. Tell it which tool and source you chose, then copy this request:

```text prompt
Help set up my chosen WireCat tool using its installation guide. Check Node.js and npm first.
Connect only my chosen account or folder; I will approve login myself.
Install the tool's instructions for my agent and prepare a small amount of data.
Test one read-only question, show its source and say what data is missing.
Explain where data stays and how to stop any background work you start.
```

The agent prepares the programs and explains the next step. You approve account access in the
service's own login flow. When the agent runs the login, it pastes the QR code or the sign-in link
into the chat for you: see [when the agent runs the login](/llms.mdx/docs/agents/content.md#agent-login). Use the tool
guide for browser consent or a bot token; do not paste passwords, tokens or authorization callbacks
into an ordinary agent chat.

For an agent without terminal access, use [MCP](/llms.mdx/docs/agents/content.md#mcp-and-skills) or [web and mobile access](/llms.mdx/docs/browser-apps/content.md).
It still needs a tool running on a computer you control.

## Set up by hand

1. **Choose the computer.** Your own Windows, macOS or Linux computer is the simplest start.
   A server is useful for recurring work, but its sessions and files live there. A laptop folder
   does not automatically exist on the server. You do not need both.
2. **Check the runtime** using the commands below. Then use the install command in your selected
   tool's guide. A command-line tool, or CLI, is a program the agent can run through a terminal.
3. **Connect your source.** Follow that guide's account or file-import path. Installing a package
   does not authorize an account. Login also does not download all its data.
4. **Connect your agent.** [Install the skill or configure MCP](/llms.mdx/docs/agents/content.md). A skill gives a
   terminal agent instructions; MCP exposes capabilities to a compatible agent.
5. **Prepare the first data and verify it.** Use the checks below before adding more sources.

## Node.js, npm and finding the command

Node.js (The program that runs tools such as tg, max and memo on your computer. Your agent can check whether it is installed and help set it up.) runs the tools; npm installs them. The supported
Node.js lines are **22.16+ within 22.x, or 24+**; tg needs **22.19+** within 22.x. Follow the chosen
tool's guide for other runtime options. In Terminal or PowerShell, run:

```sh
node --version
npm --version
```

If either is missing or the version is too old, ask your agent to prepare it:

```text
Check whether Node.js and npm are installed.
If needed, install a compatible stable version of Node.js with npm for my operating system: 22.19+ (22.x) or 24+.
Verify node --version and npm --version, then continue setting up the messenger.
```

After installing, check the tool's version in the **same environment your agent uses**. Global
installation puts commands on PATH, the folders the terminal searches. Reopen the terminal or
restart an already-running agent if it cannot find the new command.

### npm's warning about install scripts

npm 11.17 and later warn after a fresh global install of `tg`, `max` or `memo`:
`install scripts not yet covered by allowScripts`. It names the tool's own setup script and its
database driver. With npm 11 both scripts still run and the tool works, so you can ignore the
warning. npm 12 blocks them instead. Run the `npm install -g --allow-scripts=…` command that npm
prints under the warning: it allows only the packages it names, once.

### Windows and command visibility

If PowerShell blocks `npm.ps1`, use `npm.cmd`. A tool's `.cmd` launcher may also be needed; see
the Windows steps in its guide. Telegram and MAX also provide an installer that can configure
PATH and agent instructions. Do not assume that installer supports every tool. Use the exact
steps for [Telegram](/llms.mdx/docs/tg/installation/content.md), [MAX](/llms.mdx/docs/max/installation/content.md),
[Drive](/llms.mdx/docs/drive/installation/content.md) or [Zoom](/llms.mdx/docs/zm/installation/content.md).

## Check the first useful result

Ask your agent to check the account or folder you selected, retrieve a small period or import a
small folder, and answer one question with a source:

- **Messages:** show a few chats, retrieve the period you need and find a known message.
  [Telegram history](/llms.mdx/docs/tg/archive/content.md) · [MAX history](/llms.mdx/docs/max/archive/content.md).
- **Files or email:** import a chosen folder or configured mailbox and find a known phrase.
  [Memo](/llms.mdx/docs/memo/content.md) · [email](/llms.mdx/docs/mail/email/content.md).
- **Drive:** import one selected folder and open the document behind a result.
  [First Drive import](/llms.mdx/docs/drive/usage/content.md).
- **Meetings:** retrieve or import one transcript and read a cited passage.
  [First Zoom result](/llms.mdx/docs/zm/installation/content.md#check-your-first-result).

Success means the command is available, the intended source is connected, useful data is saved
and the agent can cite it. Check dates and unreadable or missing files before interpreting an empty
answer. A successful login, skill installation or MCP connection alone does not establish this.

## Data, updates and background work

Saved sessions and local data stay where the tool runs. Your agent may send retrieved content to
its AI provider, according to its settings. Review [security and data flow](/llms.mdx/docs/security/content.md) and
[permissions](/llms.mdx/docs/configuration/content.md#permissions) before allowing changes.

Installation does not start continuous syncing or auto-replies. Recurring work needs a running
process or schedule; ask your agent how to stop anything it starts. Closing its chat does not
necessarily stop a separate process. See [recurring tasks](/llms.mdx/docs/prompting/content.md#recurring-tasks).

For updates and removal, use the chosen tool's installation guide. Removing a package, deleting
local data and revoking account access are separate actions. Once your first check works,
[choose a useful next task](/llms.mdx/docs/content.md#try-a-request).
