Note: This is separate from Penny, Nickel's own built-in assistant. Penny lives in the dashboard under Penny → Chats and has its own confirmation flow. If you would like Penny enabled on your account for free, please message support@nickel.com or contact your account manager.
Before You Start
Connecting an agent takes an admin on your Nickel organization. The connection then acts with that admin's permissions, the same as if they were using the dashboard themselves.
If you see "You do not have the relevant permissions to connect apps" during sign-in, the account you signed in with is not an admin. Ask an admin on your team to connect instead.
Get Your Server Endpoint
Go to Settings → Integrations → MCP.
Under Connect a client, copy the Server endpoint:
Add it to your MCP client. The page also has a Claude Code tab with the exact command, and an Other MCP clients tab with a configuration block you can copy.
Connect Your Client
Claude (Web or Desktop)
In Claude, open Settings → Connectors.
Choose Add custom connector.
Paste https://mcp.nickel.com/mcp as the URL.
Choose Connect.
Claude Code
Run this in your terminal:
claude mcp add --transport http nickel https://mcp.nickel.com/mcp --scope user
Then run /mcp and choose nickel to start the sign-in. The --scope user part makes Nickel available in every project on your machine.
Codex
Add this to your ~/.codex/config.toml file:
[mcp_servers.nickel]
url = "https://mcp.nickel.com/mcp"
Grok
Go to grok.com/connectors.
Choose New Connector, then Custom.
Name it Nickel and paste https://mcp.nickel.com/mcp as the server URL.
Note: Custom connectors are available on Grok's paid plans.
Any Other MCP Client
Nickel works with any client that supports MCP. Most of them read a configuration file that looks like this, which you can copy straight from the Other MCP clients tab on the Settings → MCP page:
{ "mcpServers": { "nickel": { "type": "http", "url": "https://mcp.nickel.com/mcp" } } }
Check your client's own documentation for where that file lives.
Sign In and Allow Access
The first time your client connects, it opens a browser window on app.nickel.com.
Sign in to Nickel the way you normally do. If you are already signed in, Nickel skips ahead.
Nickel shows you which client is asking and which organization it will reach. Choose Allow access.
The browser hands you back to your client, which is now connected.
There are no keys to copy or store, and your client renews its own access.
Verify Your Connection
Ask your agent something about your account, for example:
Which of my invoices are overdue?
What bills are due this week?
It should answer from your live Nickel data. If it says it has no Nickel tools, reconnect and make sure you completed the Allow access step.
Tokens for Headless Agents
Some clients cannot sign in through a browser, such as a script, a scheduled job, or something running in CI. For those, create an API token on the same Settings → MCP page and send it with every request.
Important: A token acts as your organization. Create one per agent or device, so that if one needs revoking you are not cutting off everything at once. The dashboard shows the token once, so copy it then. An MCP token works only with the MCP server, and a REST API token from Settings → API does not work here.
Managing Connected Apps
Everything connected through browser sign-in is listed under Connected apps.
Note: Disconnecting an app immediately revokes its access.
Tokens for Headless Agents
Some clients cannot sign in through a browser — a script, a scheduled job, something running in CI. For those, create an API token on the same page.
Important: A token acts as your organization. Create one per agent or device, so that if one needs revoking you are not cutting off everything at once.
Approving What an Agent Asks to Do
An agent can read your Nickel data freely. Anything that would change it — depositing a check, paying a bill, creating a record — stops and waits for you.
Go to Agents → Agent requests in the main menu, or Requests from your agents.
Requests waiting on you are listed under Waiting for your approval, with a count. The menu item carries the same count as a badge, so you do not have to go looking.
Note: You only see the requests your agents made. They are not shared with the rest of your team.
Tip: After you approve a request, tell your agent to continue. Approving does not resume the agent by itself.
What a Request Looks Like
Each row tells you what the agent wants to do, in plain language, with the amount beneath it:
Pay the $3,200.00 Piedmont Hauling & Grading bill PH-4417 by ACH
Deposit a $4,820.00 check from Lakeview Commercial Group into Nickel Banking
Underneath each one, Nickel names which client asked, when it asked, and when the request expires.
Important: Requests expire. A request left undecided for about two days lapses, and the agent has to ask again. If you are not ready to approve something, that is fine — it will not sit there forever waiting to surprise you.
Deciding
One at a time: tick the request and use Approve or Decline.
Several at once: use Select all, or tick the ones you want, then Approve or Decline.
After You Decide
Approving clears the request from the list. It does not resume the agent by itself — go back to your agent and tell it to continue.
Important Reminders
An agent acts within your own permissions. It cannot do anything your role does not allow.
If your workspace uses approval rules, an action an agent sets up still goes through them, on top of your own approval here.
Tokens act as your organization. Treat one like a password, create one per agent, and revoke it the moment it is no longer needed.
Disconnecting an app cuts its access immediately.
Need help?
If you have questions, please chat with our team via the Help Center when you login to the platform.


