Connect your AI client

The Connect your AI client dialog in your APIANT account menu shows one address. Add it to your AI client as a remote MCP server. There is nothing to install: you sign in once, in your browser, the first time the client connects. The skills that teach Claude APIANT's procedures reach the client over the same connection.

Any client that takes a remote MCP server URL can connect. This page shows Claude Code, because the dialog also gives the one-line command that registers the address there.

Prerequisites

An APIANT.ai account whose role has the MCP Tools permission. New accounts have it. Without it the sign-in is refused, and only an administrator can grant it.
Claude Code, with the claude command on your PATH, or another AI client that accepts a remote MCP server URL.
A default browser on the same machine, for the sign-in.

Connect

1. Open the dialog

Sign in at app.apiant.ai.
Open the account menu (your initials and name, top right).
Select Connect your AI client. The same item is in the side panel of the Assistant page.

The dialog shows the address, a Copy button, and underneath, the command that registers it in Claude Code.

2. Copy the address

Select Copy next to the address.

Always copy it from the dialog rather than from this page. An account in the shared APIANT.ai workspace gets the address https://mcp.apiant.ai; an organization on its own APIANT domain gets an address on that domain.

3. Add it to your client

Claude Code. The dialog's second line registers the address for the current project. Open a terminal, change to the directory you want to use Claude Code in, and run it:

bash

$ cd ~/projects/acme-integrations
$ claude mcp add --transport http -s project apiant-ai <address from the dialog>

This writes a server named apiant-ai into ./.mcp.json for that directory, so it is there whenever you work there. Use --scope user instead to register it everywhere at once. Keep the name apiant-ai: the skills refer to the APIANT tools by it.

Another client. Paste the address where the client asks for a remote MCP server URL. Name the server apiant-ai where the client lets you.

4. Start Claude Code and approve the server

Run claude in the same directory, or restart Claude Code if it was already open there.
Approve the apiant-ai server when Claude Code asks about the project's MCP servers.

Run /mcp, select apiant-ai, and choose to authenticate. Claude Code also shows a notice at startup when a server needs sign-in.
In the browser page that opens, sign in with your APIANT email and password, or Continue with Google or Continue with GitHub. If your email belongs to more than one organization, pick one. If your account uses two-factor authentication, enter the 6-digit code.
On Authorize access, check the application, the permissions listed and the account under Signed in as, then select Allow.
Return to Claude Code. claude mcp list shows the server as connected.

Another client opens the same sign-in page the first time it connects.

6. Check the connection

Ask:

text

Check that you're connected to APIANT.

Where the connection lives

./.mcp.json: the apiant-ai server entry. Each directory is a separate registration: run the command again in every project directory where you want APIANT.
The sign-in token is stored by Claude Code, not in the project.

Nothing is written under ./.claude/. The skills are served over the connection and are never files in your project.

Skills

When Claude starts building, editing or testing on your account, it loads APIANT's procedures over the connection. Skills update on APIANT's side; there is nothing for you to refresh. See Skills.

Cost

Using the connection spends no assistant credits, on any plan. Your automations still use your plan's monthly tasks when they run, however you built them. See Plans, usage and billing.

Organization domains

If your organization uses its own APIANT domain, the dialog shows an address on that domain, and sign-in stays on that domain. The server name stays apiant-ai.

If the dialog says the workspace does not have a web address yet, an administrator has to set the workspace's address before an AI client can connect.

Disconnect

These are Claude Code's own steps, not APIANT's.

Run /mcp, select apiant-ai, and choose Clear authentication to remove the stored sign-in.

Remove the server registration:

bash

$ claude mcp remove apiant-ai -s project

If you connected before 2026-09-19

The earlier dialog gave a command that downloaded skill files into ./.claude/skills/ and registered the address https://mcp.apiant.ai/mcp.

The address is still correct, but the tools behind it changed. Reconnect the server so your client picks up the current tools: restart the client, or toggle the connection off and on. Until you do, a call to one of the old tool names is refused with "is not a tool on this connection".

The skill files are no longer updated. Delete the APIANT skill folders under ./.claude/skills/, and leave the rest of .claude/ alone, since nothing else in it belongs to APIANT.

Troubleshooting

Symptom	Cause and fix
`claude: command not found`	The Claude Code CLI is not on your `PATH`. Install or locate it, then run the command again.
Connect your AI client is not in the account menu	The item is hidden while you are acting as another account. Return to your own account.
`apiant-ai` shows as pending approval	Run `claude` interactively in the directory and approve it.
The server shows as needing authentication	Run `/mcp`, select `apiant-ai`, and authenticate.
Claude says it has no APIANT tools	Start Claude Code from the directory where you registered the server, and check `claude mcp list` there.
A tool call is refused with "is not a tool on this connection"	The client connected before 2026-09-19 and still expects the old tool names. Reconnect the server, as above.
Sign-in says your account does not have the MCP Tools permission	Ask your account administrator to add MCP Tools to your role.

More symptoms are covered in Troubleshooting MCP.