Agility CMS

Model content, author pages and publish to a live site from an AI agent.

Opens Agent Studio, where connecting is one click. The connector URL below works in any MCP client.

Paste it into any MCP client. Setup docs

What the Agility CMS connector does

The Agility CMS MCP server connects a headless content management system to Claude, ChatGPT and any MCP-compatible agent. It models content, authors pages, uploads media and publishes updates to a live website. Agility documents 32 tools; Anthropic's directory lists 27 and omits every publish tool. Sign-in is OAuth 2.0.

Verified connector

Listed by Anthropic as a partner connector in its Connectors Directory.

Connection checked by Agentman on .

Anthropic states this reflects the level of review a connector received, not a security audit.

Agility CMS tools (27)

  • delete_content_item
  • delete_media
  • delete_page
  • get_available_instances
  • get_component_model_details
  • get_component_models
  • get_containers
  • get_content_item
  • get_content_items
  • get_content_model_details
  • get_content_models
  • get_current_user
  • get_locales
  • get_page
  • get_page_models
  • get_sitemap
  • get_sitemaps
  • initialize_media_upload
  • list_media
  • manage_content_workflow
  • manage_page_workflow
  • reorder_page_modules
  • save_component_model
  • save_container
  • save_content_items
  • save_content_model
  • save_page

Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.

Limits

  • Anthropic's listing is five tools short, and the gap is the publish path. The directory's 27 names are a strict subset of the 32 Agility documents. Do not audit this connector from the directory entry alone — it contains no publish verb, and the connector publishes.
  • No application scopes exist. The authorization server at https://mcp.agilitycms.com advertises exactly one scope, offline_access, which governs refresh-token lifetime rather than access. The RFC 9728 resource descriptor declares no scopes_supported at all. There is no way to grant reading without also granting publishing and deleting.
  • The only real boundary is which Agility user you sign in as. Agility says so directly: your account permissions are the ceiling and the Management API refuses what that user cannot do. A broadly-privileged account grants a broadly-privileged agent.
  • The confirmation gate depends on your MCP client, not on the server. Where a client does not support MCP elicitation, Agility states the action runs and the response merely records humanConfirmed: false. Choosing "always allow" in any client also opts out, and Agility notes the server cannot override that.
  • The approval workflow is not an oversight gate against an agent. Agility documents approve, decline and request-approval as "not gated (low risk)", so an agent can request and grant approval on the same item.
  • The two vendor sources disagree about which tool publishes. The documentation page attributes publishing to manage_content_workflow; the live catalog attributes it to publish_content and says the workflow tool does not publish. The catalog and README are newer and agree.
  • We could not read tool schemas or safety annotations. The endpoint returned 401 to an anonymous request, so the destructiveHint claims are Agility's, not our observation.
  • The source is not published. The repository contains documentation only, so the permission and confirmation behaviour cannot be verified by reading code — unlike a connector whose dispatch path can be inspected.
  • The documentation page cites a repository that does not exist. It links source to github.com/agility/agility-mcp-server, which returned 404 on 2026-08-23. The real repository is github.com/agility/mcp-server. The same page's npx install for agility-mcp-server also 404s on the npm registry, so the documented local install cannot work as written.
  • Content is untrusted input and Agility says so. Its security policy names prompt injection via content items and field descriptions as structural to MCP.
  • The hosted server logs truncated payloads. Agility states request and response payloads are recorded to Application Insights in truncated form, capped at 2 KB by default, so small content values can appear in those logs, with no client-side opt-out on the hosted server.
  • No rate limits are published. Neither Agility's documentation page nor its README states a rate limit, quota or throttling behaviour for the MCP server. Treat throughput as unspecified.

Frequently asked questions

Can the Agility CMS MCP server publish content to a live website?

Yes, and that is the tool's stated purpose. Agility's live tool catalog documents publish_content as taking content items live and publish_page as taking pages live, both read on 2026-08-23. Anything published becomes visible to real visitors on your website. Neither tool appears in Anthropic's directory listing of 27 tools.

Why does Anthropic list 27 Agility CMS tools when Agility documents 32?

The directory entry is a snapshot of version 2.0.0 taken when the connector was added on 2026-06-19. Agility's README and live tool catalog both document 32 tools. The five missing names are publish_content, unpublish_content, publish_page, unpublish_page and save_page_model, so the omission falls entirely on the publishing surface.

Can the Agility CMS connector delete content or a whole page?

Yes. Three tools delete: delete_content_item, delete_page and delete_media. Agility describes each as permanently removing the record, moving it to a Deleted state. Deleting a page removes it from the sitemap, and Agility advises unpublishing a live page first. All three carry destructiveHint so interactive clients prompt before running them.

Does saving content in Agility CMS make it live immediately?

No, and Agility is explicit about it. Its documentation states that save_content_items always saves to the instance's default workflow state, typically Staging, and that a state value passed on save is not honored. Taking content live requires a separate publish call, so an ordinary save cannot reach your visitors by itself.

Can an AI agent approve its own Agility CMS content for publishing?

Yes. manage_content_workflow and manage_page_workflow perform approve, decline and request-approval actions, and Agility's documentation states approve and decline are not gated because they are low risk. An agent holding a permitted account can therefore request approval and then approve the same item without a person intervening.

What OAuth permissions does the Agility CMS MCP server request?

Only offline_access, which controls refresh-token lifetime rather than access. Agility's authorization server advertised that single scope on 2026-08-23 and the RFC 9728 resource descriptor names no scopes at all. Nothing at the consent screen separates reading a content item from deleting one or publishing it.

How do I limit what an AI agent can change in Agility CMS?

Sign in as a restricted Agility user. Agility states your account permissions are the ceiling, that every call runs as you, and that the Management API refuses anything you personally cannot do. Because no OAuth scope narrows anything, the Agility user account you authenticate with is the only real boundary available.

Is the Agility CMS MCP server open source?

No. The public repository at github.com/agility/mcp-server contains only documentation, a licence and a security policy. Its README states the server's source is maintained by Agility CMS and is not currently published. So the confirmation and permission behaviour described in the docs cannot be verified by reading code.

Sources

Use it in an agent

Put Agility CMS to work.

Connect Agility CMS once and your agents call these tools on their own: on a schedule, in a workflow, with nobody at the keyboard.

Call (650) 285-1019Our AI receptionist answers.