# MCP server

Footprint's MCP server connects the AI tools your team already uses to your Footprint organization. Once connected, you can ask Claude, Claude Code, Cursor or Codex to look up a user, explain why an onboarding failed, count risk signals, or draft a playbook change, and it works with your own Footprint permissions.

The server URL is:

```text
https://api.onefootprint.com/mcp
```

## Before you start

* **Your role needs to have the MCP permission.** The **Admin** role always has it. For any other role, an admin turns on **MCP** for that role under **Settings** > **Members and roles** > **Roles**. MCP only lets the role connect: each action still needs the role's usual permission, such as decrypting a field or editing playbooks.
* **Your organization has live mode.** Other organizations can ask [support@onefootprint.com](mailto:support@onefootprint.com) for access.
* **An AI tool that supports remote MCP servers with sign-in**, such as one of the four below.

## Connect your AI tool

Add the server URL to your tool, then sign in to Footprint when it asks. Each tool keeps its own connection, so connect each one you use.

### Claude

Claude Desktop uses the same connectors as claude.ai. Anthropic's guide: [Custom connectors using remote MCP](https://support.claude.com/en/articles/11175166-getting-started-with-custom-connectors-using-remote-mcp).

On Team and Enterprise plans, members may not be able to add custom connectors, depending on the plan and admin settings. If **Add custom connector** is greyed out, ask your Claude organization owner to add Footprint under **Organization settings** > **Connectors**, then select **Connect** on it under **Customize** > **Connectors**.

1. Open **Customize** > **Connectors**, select **Add**, then **Add custom connector**.
2. Name it `Footprint`, paste the server URL, and follow the prompts to add it.
3. Select **Connect** next to the connector and sign in to Footprint.

### Claude Code

Add the server from your terminal:

```bash
claude mcp add --transport http --scope user footprint https://api.onefootprint.com/mcp
```

`--scope user` makes it available in every project; leave it out to add it to the current project only. Then run `/mcp` inside Claude Code, select **footprint**, and sign in to Footprint in your browser. Anthropic's guide: [Connect Claude Code to tools via MCP](https://code.claude.com/docs/en/mcp).

### Cursor

Add the server to `~/.cursor/mcp.json`, or to `.cursor/mcp.json` in a project:

```json
{
  "mcpServers": {
    "footprint": {
      "url": "https://api.onefootprint.com/mcp"
    }
  }
}
```

Then find **footprint** under **Customize** in Cursor's sidebar and select its sign-in prompt to sign in to Footprint. Cursor's guide: [Model Context Protocol](https://cursor.com/docs/context/mcp).

### Codex

Add the server and sign in from your terminal:

```bash
codex mcp add footprint --url https://api.onefootprint.com/mcp
codex mcp login footprint
```

In the Codex app, open **Settings** > **MCP servers** instead (in the IDE extension, the gear menu > **MCP servers**), select **Add server**, choose **Streamable HTTP**, enter the server URL, save and restart, then select **Authenticate**. OpenAI's guide: [MCP in Codex](https://learn.chatgpt.com/docs/extend/mcp?surface=cli).

## Sign in and choose an organization

When your AI tool asks you to sign in, it opens Footprint in your browser:

1. Sign in to Footprint as you do for the dashboard.
2. On **Add a connector**, check the account you're signed in as, then select **Choose organization**.
3. Pick the organization this tool should work in, and approve access.

The list shows each organization where your role has the MCP permission. If none is listed, ask an admin to add MCP to your role, or sign in with the method your organization requires, such as SSO. A connection to a live organization can use both live and sandbox data, so say which one in your request. To work in a different organization, connect again and pick that one.

If you pick a sandbox-only organization, sign-in can succeed while Footprint refuses every request, and your AI tool may only say it couldn't connect. Connect again and pick a live organization, or ask [support@onefootprint.com](mailto:support@onefootprint.com) for access.

Your AI tool renews the connection on its own, so you normally stay signed in until you disconnect. If a sandbox-only organization later goes live, its connections stop working: remove Footprint from your AI tool and connect again.

## What you can do

Your AI tool can work across these areas of Footprint:

* Users, businesses and their onboardings
* Reviews
* Playbooks
* Agents and their investigations
* Risk signals

Everything runs with your role's permissions, so the AI tool can only do what you can do in the dashboard.

## Permissions and audit

Footprint checks your current role and membership on every request. If an admin changes your role or removes you from the organization, the change applies to your AI tool right away.

Footprint records actions taken through MCP under your name, the same way it records them from the dashboard. Decrypting vault fields needs a reason, which is recorded with it.

The MCP server acts with your permissions. Data in Footprint, such as names, documents and notes, comes from your users, and data from other tools can come from anywhere. Either can contain instructions written to steer an AI tool. Review each change, decrypt or playbook run before you approve it, and don't auto-approve Footprint's write tools. Take extra care when the same AI tool can also send data elsewhere, such as by email or to the web. A role with fewer permissions limits what a connection can do.

## Disconnect

To disconnect an AI tool, remove Footprint from it:

* **Claude:** in **Customize** > **Connectors**, select **Remove** from the Footprint connector's menu. On Team and Enterprise plans, an owner can also remove it under **Organization settings** > **Connectors**.
* **Claude Code:** run `claude mcp logout footprint`, then `claude mcp remove footprint`.
* **Cursor:** remove the server under **Customize**, or delete the `footprint` entry from `mcp.json`.
* **Codex:** run `codex mcp logout footprint`, then `codex mcp remove footprint`.

Removing Footprint from a tool disconnects that tool, but doesn't revoke its sign-in on Footprint's side. If you think a device or its credentials are compromised, ask an admin to remove the MCP permission from your role. That stops every connected tool on its next request.

## Feedback

To report what works, what breaks, or what you'd like it to do, email [support@onefootprint.com](mailto:support@onefootprint.com), or tell your Footprint contact in your shared Slack channel.