Connect Claude Code, Codex, or Cursor to Agent Loadout

Every client that talks to Agent Loadout uses the same server URL and the same token. You connect once, the agent’s 33 MCP tools become available in your client, and swapping to a different client means pointing it at the same URL — no new agent, no new credentials. MCP server URL

https://agent-loadout.com/api/mcp

The server uses OAuth 2.1 with PKCE. Clients that support browser-based OAuth (Claude Code, Cursor, ChatGPT) handle authentication automatically. For programmatic access, use a token you issue in the agent’s Tokens tab.

Claude Code
Codex CLI
Cursor
ChatGPT
REST API

Run the following command in your terminal. Claude Code opens your browser so you can sign in, choose an agent, and approve its permissions.

Connect Claude Code

claude mcp add --transport http agent-loadout https://agent-loadout.com/api/mcp

After you approve access in the browser:

Return to your terminal — Claude Code reports 33 tools available.
Open a new chat session and enable Agent Loadout in the tools menu if prompted.
Ask Claude Code which agent it is connected to, and it will call get_agent to confirm.

No local installation or API key is required. To disconnect, revoke the token in the agent’s Tokens tab and remove agent-loadout from Claude Code’s MCP server list.

Add Agent Loadout as an HTTP MCP server in your Codex configuration file. The exact path depends on your Codex installation — typically ~/.codex/config.json or codex.yaml.

codex.json

{
  "mcpServers": {
    "agent-loadout": {
      "type": "http",
      "url": "https://agent-loadout.com/api/mcp"
    }
  }
}

On first use, Codex will prompt you to authenticate. Sign in to Agent Loadout in the browser, choose your agent, and approve its permissions. Codex stores the resulting token and refreshes it automatically.

If Codex does not prompt for authentication automatically, generate a token in the agent’s Tokens tab and supply it via the AGENT_LOADOUT_TOKEN environment variable instead.

Add Agent Loadout in Cursor’s MCP settings:

Open Cursor Settings (⌘, on macOS or File → Preferences → Settings on Windows/Linux).
Navigate to the MCP section.
Choose Add new MCP server.
Enter the name agent-loadout and the server URL https://agent-loadout.com/api/mcp.
Save the settings.

Cursor opens your browser to complete authentication. Sign in to Agent Loadout, select the agent you want Cursor to use, and approve the requested permissions. The tools become available in Cursor’s chat panel automatically.

ChatGPT supports Agent Loadout via a custom connector. A directory listing is not yet published, so you need to add it manually.

Custom connectors require Developer mode. Availability depends on your ChatGPT account and workspace policy.

In ChatGPT, open Settings → Security and login and enable Developer mode.
Open Plugins, choose the plus button, and name the connection Agent Loadout.
Enter the server URL https://agent-loadout.com/api/mcp and choose OAuth when asked for the authentication method.
Complete sign-in to Agent Loadout in the browser and review the requested access.
In a new chat, open the tools menu and add Agent Loadout to the conversation.

After setup, enable Agent Loadout in any chat session from the tools menu to make the agent’s capabilities available.

Use the same token for direct REST calls. Pass it as a Bearer token in the Authorization header. The base URL is https://agent-loadout.com/api/v1.

Verify your identity

curl https://agent-loadout.com/api/v1/me \
  -H "Authorization: Bearer $AGENT_LOADOUT_TOKEN"

List inboxes

curl https://agent-loadout.com/api/v1/inboxes \
  -H "Authorization: Bearer $AGENT_LOADOUT_TOKEN"

Send a message

curl -X POST https://agent-loadout.com/api/v1/inboxes/<INBOX_ID>/messages \
  -H "Authorization: Bearer $AGENT_LOADOUT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "to": ["person@example.com"],
    "subject": "Hello from my agent",
    "text": "Sent by my agent.",
    "idempotency_key": "hello-1"
  }'

The full OpenAPI spec is at https://agent-loadout.com/api/v1/openapi.json. Errors return JSON with a matching HTTP status code:

Error response

{
  "error": {
    "code": "forbidden",
    "message": "…",
    "hint": "…"
  }
}

Status codes: 400 bad request, 401 unauthenticated, 403 forbidden, 404 not found, 409 conflict, 429 rate limited, 402 quota or budget reached.

Try it after connecting

Once your client is connected, try these prompts in any chat session. Tool access follows the agent’s granted permissions and equipped resources.

Which Agent Loadout agent am I connected to, and what is its email address?

Summarize the new messages in my Agent Loadout inbox.

Draft a reply to the newest email. Save it without sending it.

List the accounts in my agent's vault without showing their passwords.

Show my agent's payment cards and spending limits without revealing card details.

List my agent's machines and tell me which are running.

Email, credentials, cards, and machines all share the same connection. Access to a credential’s value must be enabled for the agent. Cards and machines must also be available in your workspace. Sending and running machines follow your account’s plan limits.

Troubleshooting

The connector option is missing in my client

Some accounts require Developer mode or a workspace administrator to enable custom connectors. Check the official setup guide linked in your client’s own documentation. For ChatGPT specifically, Developer mode must be enabled under Settings → Security and login before the custom connector UI appears.

I signed in, but my chat cannot see any tools

Complete the access-approval step in the browser first, then return to your client. If the tools still do not appear:

Start a new conversation — some clients require a fresh session before tools become visible.
Look for a Refresh tools or Reconnect option in the client’s MCP or connector settings.
Make sure you approved permissions for the correct agent during the OAuth flow. If you are unsure, disconnect and reconnect to choose again.

A tool returns 'permission denied' or 'resource not found'

This means the token used for the connection does not include the required capability, or the resource (inbox, credential, machine) is not equipped on the selected agent. To fix it:

Open the agent’s Tokens tab and check which capabilities the token carries.
Open the agent’s Equipment tab and confirm the inbox, vault entries, or machines are present.
If you need to add capabilities, issue a new token with the required scopes and reconnect — or disconnect and reconnect to choose a different agent.

How do I disconnect Agent Loadout from a client

In the Agent Loadout dashboard, open the agent’s Tokens tab.
Find the token issued for that client and choose Revoke. This immediately stops all access from that token.
In your client’s MCP or connectors settings, remove the Agent Loadout server entry.

You can reconnect at any time by pointing the client at https://agent-loadout.com/api/mcp again and completing a new OAuth flow.

Get Started

Core Concepts

Guides

Account & Billing

Connect Claude Code, Codex, or Cursor to Agent Loadout

Try it after connecting

Troubleshooting

​Try it after connecting

​Troubleshooting

Try it after connecting

Troubleshooting