Skip to content

AI agents (MCP)

You can manage your apps from an AI agent that supports MCP (Model Context Protocol) without opening the backoffice. You ask for what you need in natural language and the agent does it in your app, with your user and with your same permissions.

RequirementDetails
PlanThe AI agents (MCP) feature is part of the app’s plan. Its availability depends on the corresponding contract.
Access in the appAI agent access is enabled on every app. The Reskyt team can disable it so that no agent can operate on that app.
UserA backoffice user with access to the app.
Compatible agentClaude, ChatGPT, Cursor, VS Code or an MCP client that runs on your computer, such as Claude Code.

Only those applications and clients running on your own computer are accepted. If you need to connect another one, ask the Reskyt team.

The quickest way to connect Claude is from the backoffice. Open Profile in the top bar, or go directly to admin.reskyt.com/customer/profile. In the AI agents card, press Connect to Claude: Claude opens with the Reskyt connector already filled in. Confirm it and follow steps 2 to 4 below.

The same card shows the server address with a button to copy it, for use in other applications. It only appears if at least one of your apps has AI agent access available.

The server address is:

https://mcp.reskyt.com/mcp
  1. Add the connector in your agent. Use the address above as a custom connector or MCP server.

  2. Confirm the connection. A Reskyt page opens showing which application wants to connect. Check that you started the connection yourself and press Continue.

  3. Sign in. Sign in with your Reskyt user, as in the backoffice, also with Google if that is how you usually sign in.

  4. Try it. You return to the agent, now connected. Ask it, for example, “which apps can I manage?”.

Use the Connect to Claude button in your profile, or paste the server address in Settings → Connectors → Add custom connector.

To disconnect it, remove the connector in your agent.

The agent can do the same as you in the backoffice, and nothing more. If your role does not allow an action, the agent cannot do it either.

AreaExamples of what you can ask
Content“Create a Black Friday landing page, add it to the menu and publish it.” It also edits pages, pop-ups, the menu, the CSS and the JavaScript, and restores revisions.
Theme“Apply our brand colours and the font I am sending you.” It changes the theme’s colours, typography and corners.
Push notifications“Send a push to Android users in France on Friday at 10:00.” It creates templates, calculates audiences, sends, schedules and enables automations.
Analytics“How many iOS devices do we have in Spain?”, “how did yesterday’s push perform?”
Quality“Why did the last QA fail?”, “Create a test that checks the cart opens and run it.” It also reviews a page before publishing it.
Variables“The cart badge does not update: what is wrong with cart_count?” It tells you whether each variable updates live and, if not, why and how to fix it.
Reskyt Play“Reorder the plays and tell me which one has the most views.”
Users“Invite our new marketing manager and give them access to the Spain zone.”

The agent checks the platform documentation (menus, components, variables and pop-ups) before creating content.

When it saves content, the agent creates a revision. That change does not reach the app’s users until it is published to an environment; you can ask it to publish in the same step or do it yourself from the backoffice.

Any change made by the agent can be recovered from Revisions, like any other change made in the backoffice.

Your permissions, no more

The agent acts with your user through the same API the backoffice uses. Permissions, app access and zones apply the same way.

Sign-in with consent

The connection uses OAuth 2.1 with PKCE. Every new connection asks for your confirmation and your Reskyt sign-in.

No exposed credentials

Tokens, passwords and API keys are never returned to the agent: they are managed from the backoffice.

Traceability

Revisions, sendings and the change log show your email followed by (via MCP), so you can tell what the agent did.

Text the agent reads from the app that was written by users or third parties is handed to it marked as untrusted content, so it treats it as data and not as instructions. Reskyt records the use of the MCP server: which action was performed, in which app, from which agent and with what result.