What is your developer-dependent CMS actually costing you? Calculate your costs and get a full report
What is your developer-dependent CMS actually costing you? Calculate your costs and get a full report
Security and Compliance
For admins and security reviewers: what each Agility AI surface can read and change, where your content goes, and what stays in Staging.
This page is for admins, security reviewers and data protection teams who need to know what data moves where when people use AI with Agility CMS. It covers each AI surface Agility documents today and what each one can read and change.
It is about data handling. For how to control who can do what with AI, and how to roll it out, see Governing AI Access to Agility CMS.
| Surface | What it does | Reads | Writes | How it is accessed |
|---|---|---|---|---|
| Agility CMS MCP Server | Lets AI assistants such as Claude, ChatGPT, GitHub Copilot and Cursor work in your instances. | Instances, models, content, pages, sitemaps and media the user can access. | Models, containers, content, pages and media, plus workflow, publish and delete, within the user's permissions. | OAuth 2.0 sign-in by each user, at https://mcp.agilitycms.com/api/mcp. |
| Agility Knowledgebase MCP Server | Lets AI assistants search and read the public Agility documentation. | Public documentation only. | Nothing. It is read-only. | No sign-in. |
| DeepL app | Translates content with DeepL, an AI translation service: fields an editor picks in the DeepL panel, or items and pages as they are copied into another locale. | The fields an editor selects, or the text of the items and pages being copied with Translate Content. | Panel: translated text into the editor, kept only when the editor saves. Copy: translated drafts in the target locale, in Staging. | Installed on an instance by an administrator, with your organization's own DeepL API key. |
The MCP server calls the Agility Management API on the user's behalf, and the Management API checks that user's permissions on every call. An AI assistant can do no more through the MCP server than the person could do in the Agility app (Governing AI Access).
| Tool area | Reads | Changes |
|---|---|---|
| Discovery | The instances the user can access, the signed-in user's identity, locales and containers. | Nothing. |
| Models and schema | Content, component and page models and their fields. | Creates or updates models and containers. |
| Content | Content items and lists. | Saves content items; publishes, unpublishes, approves, declines or requests approval; deletes items. |
| Pages and sitemaps | Sitemaps, pages and page models. | Saves pages, reorders components, runs page workflow, deletes pages. |
| Media | The media library list and asset URLs. | Uploads and deletes assets. |
The current list of tools is published at mcp.agilitycms.com/tools.
Because permissions are checked on every call, the way to cut off what someone can do through AI is the same as in the app: change their role, or remove them from the instance in Settings > User Access.
When an AI saves content through the MCP server, the item lands in Staging. Saving changes to a published item creates a new Staging version while the live version stays as it is. Publishing is a separate action that needs the Publish permission, and approval workflows apply to AI-made changes the same as to changes made by hand (Governing AI Access).
Your organization chooses how much autonomy AI gets for each kind of content: Assist (AI saves to Staging, a person publishes), Approve (AI requests approval, an approver signs off) or Autonomous (a dedicated automation user publishes after checks). See Levels of autonomy.
Unpublishing and deleting are annotated as destructive, so MCP clients prompt before running them, and clients that support MCP elicitation show a confirmation form from the server. A user can turn client prompts off, so the dependable control is the user's role (Governing AI Access).
Because the MCP server acts as the signed-in user, a change made through an AI assistant is recorded against that person in the item's version history, the same as a change they made by hand (Governing AI Access). Run the read-only get_current_user tool to confirm which user an AI client is acting as.
The Knowledgebase MCP server searches and reads the public Agility documentation. It needs no sign-in and has no access to your instances or content. It is deployed on Vercel. Its search runs on Azure AI Search, and falls back to the same Algolia index that powers the search bar on this documentation site (Agility Knowledgebase MCP Server).
The only data you send it is what your AI assistant puts in a search query or document request. Avoid putting confidential information in those queries.
The DeepL app translates text with DeepL using your organization's own DeepL API key, so DeepL processes the text under your DeepL account and the terms of your DeepL plan (DeepL).
The full rollout guidance is in Governing AI Access to Agility CMS.
For anything else your review needs, such as token lifetimes, MCP hosting and logging, subprocessors, whether content is used to train AI models, or whether MCP can be turned off for your organization, contact your account team or support@agilitycms.com. For the AI client your team chooses, such as Claude or ChatGPT, training use is set by that provider's terms.
