Connect Claude Code

Connect Claude Code to Agent Vault so it can check out and sign in for you, using only mock tokens. It takes about 2 minutes: one command adds the server, and you sign in once with /mcp.

Copy the line for Claude Code from your dashboard, then add the server and sign in.

Before you start

An Agent Vault account, with a card if your agent will check out. New here? The Quickstart sets one up in about 5 minutes.
Claude Code, installed and signed in.

Paste one line
Set it up by hand

Paste this into Claude Code

Read https://mcp.agentvault.io/skill.md and follow it to connect Agent Vault.

Claude Code reads the connect skill and runs the command that adds the server.

Run /mcp in Claude Code, choose agent-vault and sign in on the Agent Vault sign-in page that opens, with your Agent Vault account.

Add the server

claude mcp add --transport http agent-vault https://mcp.agentvault.io/mcp

Claude Code answers Added HTTP MCP server agent-vault, and claude mcp list shows it as Needs authentication until you sign in. The command adds the server for the project you run it in. To use it in every project, add --scope user after --transport http.

Run /mcp in Claude Code and choose agent-vault, or run this in a terminal:

claude mcp login agent-vault

Sign in on the Agent Vault sign-in page that opens.If Claude Code runs on a server or over SSH, run this in your own terminal on that machine instead, open the link it prints and sign in:

claude mcp login agent-vault --no-browser

Your browser then fails to load an address starting with http://127.0.0.1. That is expected: paste the whole address into the terminal when it asks.

Add it in .mcp.json instead

Add this to .mcp.json in your project, keeping everything already in the file:

.mcp.json

{
  "mcpServers": {
    "agent-vault": {
      "type": "http",
      "url": "https://mcp.agentvault.io/mcp"
    }
  }
}

Choose what it may use

A new connection starts with nothing granted, so your agent can’t use a card or a login yet. Until you choose, its first tool call answers:

Choose what this app may use in Agent Vault before it can act: open the connection's page in the dashboard.

The connection itself appears in the dashboard only once your agent makes its first tool call, so make that call first.

Make a first call

Ask your agent to call get_token_catalog. It answers with the message above, and the call creates the connection.

Open Connected apps

In the dashboard sidebar, under Vault, click Connected apps. The new connection’s row shows Choose what it may use.

Pick its cards and logins

Click Choose access. Under Cards and Logins, pick All, or keep Only selected and tick the ones this agent may use.

Check the choices, then click Save access

Your first connection’s dialog opens with nothing ticked. After that, a new connection’s dialog opens with the choices of the connection you used most recently, so saving it unchanged grants the same cards and logins, and the same three switches: Let this app create and change its own logins, Ask me before every purchase from this app and Fill a test card instead of my card. Check all of them for this agent before you click Save access.

You can change or revoke the connection on the same page at any time. See Connected apps.

Let the tools run without asking

Claude Code asks before each MCP tool call unless a rule allows it, and in auto mode it can refuse a call outright, mid-checkout. Add this rule to the permissions in ~/.claude/settings.json, keeping any rules already there:

~/.claude/settings.json

{
  "permissions": {
    "allow": ["mcp__agent-vault"]
  }
}

mcp__agent-vault matches every tool the agent-vault server exposes and nothing else, so it loosens nothing for other servers or shell commands. Restart Claude Code after adding it.

Check it works

Ask Claude Code to call the Agent Vault tool get_token_catalog. It answers with the mock tokens for the cards and logins you allowed, such as av_card_1_number_u7x2k. Your real card number never reaches Claude Code.

Troubleshooting

The server is missing in another project

The add command saves it for one project. Run it again with --scope user to add it for every project.

The first tool call says to choose what this app may use

The connection has nothing granted yet. Follow Choose what it may use, then ask again.

For Claude Code, the dashboard recommends an API key over signing in: in testing, sign-in was slower or failed more often. Its Claude Code setup leads with one command that runs the local server with your key. On Overview, open Connect another agent and pick Claude Code, or see MCP setup.

Your first checkout

Have Claude Code buy something, start to finish.

Tool reference

Every tool your agent now has.

Connect your agent

Connect Claude Code

Before you start