ROOTKey MCP

Prompt it.
ROOTKey does it.

Connect the AI you already use to ROOTKey. Ask in plain language and it adds files, creates vaults and tables, integrates ROOTKey into your code and reads audit trails, with your permissions and never more.

Claude · ChatGPT · Claude Code · GitHub Copilot · OpenAI Codex · Cursor, Gemini CLI and Windsurf through the local server

PermissionWrite
Every changeNames its organisation
01 · Vaults and files

“Create a vault and
add these contracts.”

Your AI creates the vault in the organisation you chose and adds the files. Each one is anchored in ROOTKey and can be validated later.

SchemaYour own columns
02 · Tables

“Create a table for
supplier certificates.”

With the columns you describe: supplier, certificate number, expiry date. Records are written and versioned, so every change keeps its history.

ReadsYour code · the ROOTKey docs
03 · Your code

“Audit this system.
Propose how to install ROOTKey.”

A coding agent reads your existing code and the ROOTKey documentation, proposes the best way to install ROOTKey and integrates it into your code and platforms.

PermissionRead
ShowsHistory · validations
04 · Audit trail

“Who changed this record,
and when?”

Your AI reads the audit trail and the history of the file or record, in the conversation where the question came up.

Tools for vaults, files, tables, records, audit trails and the trash

Compatible

Works with the AI
you already use.

No new interface to learn. ROOTKey comes to the assistant and the editor your team already works in.

Claude, ChatGPT, Claude Code, GitHub Copilot in VS Code and OpenAI Codex connect to the hosted ROOTKey connector. Cursor, Gemini CLI and Windsurf use the local ROOTKey server. Grok has not been tested yet, and the Gemini app cannot connect yet.

Control

Your permissions.
Never more.

  • It acts as youA connected app never has more access than you have in the ROOTKey app.
  • Your organisation decidesAgents are read-only until an admin allows full access.
  • Restricted stays closedRestricted vaults and tables stay closed unless your organisation allows a time-limited opening that you approve.

Connect your AI

Pick your AI.
Follow the steps.

Choose the assistant or the coding agent you use, and copy what it needs.

No ROOTKey account yet?Create a free account first

Chat apps

Add ROOTKey as a connector

Before you start: custom connectors work on Free (one custom connector), Pro, Max, Team and Enterprise. On Team, only Owners and Primary Owners add it; on Enterprise, so can someone whose custom role includes Manage access to Libraries. Members then click Connect on it. Adding from the mobile apps is in beta, so use the web or Claude Desktop.

  1. Open ConnectorsOn claude.ai or in Claude Desktop, open Customize > Connectors. On Team or Enterprise, the Owner opens Organization settings > Connectors.
  2. Add a custom connectorClick + Add > Add custom connector (some accounts show Add > Custom > Web). A Team or Enterprise Owner, in Organization settings > Connectors, clicks Add > Custom and chooses Web if asked. Name it ROOTKey and enter this address:https://mcp.rootkey.ai/mcp
  3. Choose the sign-in settingsUnder Authentication, choose Sign in now or Sign in when needed, not No sign-in. Under OAuth client, choose Use Claude's published identity (Recommended). Leave Request headers empty and click Add. These settings cannot be changed later: to change them, remove the connector and add it again. If your dialog is a single screen with Advanced settings, enter the name and address, leave Advanced settings empty and click Add.
  4. Connect and sign in to ROOTKeyClick Connect and sign in to ROOTKey. On the consent screen, choose the permissions (Read, Write, Delete), Stay connected and the organisations, then Allow access. No token to create or paste.
  5. Turn it on in a chatIn the message box, select + > Connectors and switch on ROOTKey. When Claude asks to use a tool, choose Allow once or Always allow. You can change tool permissions (Always allow, Needs approval, Blocked) later under Customize > Connectors > ROOTKey. On Team and Enterprise, Owners can limit them for the whole organisation, and you cannot override those limits.
  6. Disconnect when doneIn ROOTKey, open Account Settings > Connected apps, open the app’s actions menu and choose Disconnect app. Then remove the connector in Claude (Customize > Connectors > ROOTKey, three-dot menu, Remove). An Owner removes an organisation connector under Organization settings > Connectors.

Steps checked against the official documentation (Claude Help Center, Claude Docs), October 2026.

Then ask

List my ROOTKey organizations and vaults.

Agents can only read until a ROOTKey organisation admin allows full access. The hosted connector uses live data only.

Coding agents

Give your agent one prompt

Or let your agent set up the local serverPaste this into your coding agent

Connect this project to ROOTKey: read https://docs.rootkey.ai/agent-setup/prompt and follow its steps to install and configure the official ROOTKey MCP server (@rootkey/mcp). Then audit this repository and propose the best way to integrate ROOTKey.

It needs a ROOTKey personal access token. Create one in the app (Account Settings > Personal access tokens) and set it as ROOTKEY_TOKEN in your environment. Never paste the token into the chat. Agents can only read until a ROOTKey organisation admin allows full access. The hosted server uses live data only; the local server can use test data with ROOTKEY_TEST_MODE=true.

Create a free account firstRead the connector guide

Governance and security

Your organisation
decides.

A connected app acts as you, and each organisation decides what agents may do and what they can reach.

Identity

It acts as you

A connected app never has more access than you have in the ROOTKey app.

Default

Read-only until an admin allows more

Each organisation decides whether agents may act. Until an admin allows full access, they can only read.

Controls

Set by your organisation

Require two-factor authentication, block a member’s agents, restrict access to allowed apps (up to 50) and see agent activity.

Restricted

Restricted vaults and tables stay closed

Agents cannot reach them. If your organisation allows it, you can open one vault or table to one connection for a few minutes, approved in ROOTKey with a fresh second factor. The approval covers reading and writing, never a delete.

Trace

Every change names its organisation

Each change an agent makes reports the organisation it acted in.

Recovery

Recovery points are listed, not managed

Agents can list recovery points. Creating, downloading or deleting one stays in the app. They can queue a file audit report, which arrives in the dashboard, and restore deleted files, records and vaults from the trash, never restricted ones.

Startup program

Member of the
Anthropic startup program.

ROOTKey is part of the startup program of Anthropic, the company that makes Claude.

Anthropic Startup program member

MCP FAQ

Questions before
you connect

Do I need to install anything?

Not in a chat app: add https://mcp.rootkey.ai/mcp as a custom connector and sign in to ROOTKey, with no token to create or paste. A coding agent runs the official ROOTKey MCP server (npx -y @rootkey/mcp, Node.js 20 or later) with a personal access token.

Which AI assistants can I use?

With the hosted ROOTKey connector: Claude, ChatGPT, Claude Code, GitHub Copilot in VS Code and OpenAI Codex. With the local ROOTKey server and a personal access token: Cursor, Gemini CLI, Windsurf (Devin Local) and other clients that run local MCP servers. Grok has not been tested with ROOTKey yet, and the Gemini app cannot connect to the hosted connector yet. The guide for each AI above says which way applies.

Can the AI see or do more than I can?

No. A connected app acts as you and never has more access than you have in the ROOTKey app. In a chat app you choose the organisations and permissions on the consent screen; with the local server, the scopes of your personal access token.

Can our organisation control what agents do?

Yes. Each organisation decides whether agents may act, and they are read-only until an admin allows full access. An organisation can require two-factor authentication, block a member’s agents, restrict access to up to 50 allowed apps and see agent activity.

Where do I get a personal access token?

In the ROOTKey app, under Account Settings > Personal access tokens. Set it as ROOTKEY_TOKEN in your environment and never paste it into the chat. Chat apps with the custom connector need no token.

Can I try it without touching live data?

The hosted connector works on live data only. The local server works on your organisation’s separate test data when you set ROOTKEY_TEST_MODE=true.

ROOTKey MCP

Connect your AI
to ROOTKey.

Add the connector, sign in and choose what your assistant may do. Then ask.