One URL

In May, I wrote that the CMS was becoming the surface agents interact with. In September, it became a product category.

What Shipped

Adobe's July 2026 AEM release included a unified MCP server. One URL. Configure it in any MCP-compatible agent — Claude, ChatGPT, Copilot — and the agent can query content models, generate fragments, manage assets, and deploy pages. A separate Document Authoring MCP server enables conversational content creation. A Governance Agent enforces brand and compliance policies using A2A (Agent-to-Agent) protocol. Three MCP surfaces on one CMS.

On September 1, Salesforce completed its acquisition of Contentful for $1.5 billion. The integration thesis: Agentforce agents query, assemble, and deliver content dynamically without manual publishing. Contentful's structured content architecture becomes the substrate that Agentforce agents compose against. Not a CMS with AI features. An agent platform with a content layer.

Between those two, the rest of the market moved:

Seven vendors. Seven MCP implementations. One proprietary CMS serving 175 million monthly users across 40 brands with zero protocol surface.

What the CMS MCP Server Actually Does

This is not a marketing feature. It is an architectural primitive. When a CMS ships an MCP server, it means:

1. External agents can discover content structure. An AI agent queries the CMS and learns what content types exist, what fields they have, how they relate. It does not need to be trained on the schema. The schema describes itself.
2. External agents can query content. Not just search — semantic queries, filtered by metadata, scoped by permissions. An agent looking for "recipes with five ingredients that take under 30 minutes" gets structured results, not a search results page.
3. External agents can create and modify content. Within role-based permissions. Brightspot's implementation scopes agent access identically to human users — same roles, same site restrictions, same content type limits.
4. External agents can orchestrate workflows. Adobe's Governance Agent uses A2A to enforce compliance policies. Brightspot connects to analytics, marketing automation, and DAM systems. The CMS becomes a coordination hub, not just a content store.

WordPress VIP's single endpoint — api.wpvip.com/mcp — replaces what previously required separate API keys for each integration. One URL, 57 tools, governed by the same permission model WordPress has enforced for twenty years.

Adobe did this with a single configurable URL. That is what "one URL" means. The entire content platform, addressable by any agent, through one protocol endpoint.

The Prediction That Became a Product

On May 30, I published The Agent Surface. The thesis: CMS platforms were entering a third era. Era 1 was monolithic (content and presentation coupled). Era 2 was headless (content decoupled from presentation via API). Era 3 was agentic (content surfaces itself to AI agents via protocol).

On June 12, The Die-Off mapped CMS vendors that were surviving the transition. Brightspot's MCP implementation was a signal. The finding: CMS platforms that built protocol surfaces survived. Those that built only distribution infrastructure did not.

Four months later, this is not a prediction. It is an industry-wide product category. "Every headless CMS will need an MCP server" is now an analyst consensus position. Acquia calls it "agentic content intelligence." FocusReactive calls it "agentic CMS." The name does not matter. The architecture does: content platforms that cannot be addressed by agents become invisible to agents.

Two Gaps, One CIO

Here is what connects this to everything else I have been writing about People Inc.

D/Cipher has a protocol gap in the ad tech layer. It has intelligence — 100 million scored connections, three-layer enrichment pipeline, a $4 CPM spread — but no MCP server, no AAMP registration, no way for buyer agents to discover its contextual capabilities. I have written extensively about this.

Now the same gap is forming in the content layer. Graphene-Selene has intelligence — knowledge graphs, entity relationships, content services across 40 brands, GraphQL APIs, document lifecycle hooks. But no MCP server. No agent surface. No way for external AI agents to discover what content exists, query it semantically, or compose against the content model.

This matters more than the D/Cipher gap, and here is why: the ad tech protocol gap threatens the session-based half of digital revenue ($165M, flat). The CMS protocol gap threatens the content production pipeline that feeds all revenue. If Graphene-Selene cannot be addressed by agents, then the "50% more content at same cost" AI productivity claim that Vogel made at Goldman Communacopia depends entirely on internal AI tools. It cannot compound with the external agent ecosystem that every competitor CMS is joining.

What Competitors Can Do That Graphene-Selene Cannot

A media company running Adobe AEM can tell an AI agent: "Find all recipes published this month with ratings above 4.5 and fewer than 8 ingredients, then create a curated collection page." The agent queries the MCP server, discovers the content model, filters by the fields, and assembles the collection. One URL. No custom integration.

A media company running Brightspot can give an AI agent the same permissions as a junior editor. The agent creates content, triggers review workflows, publishes to staging — all within the existing editorial governance. The same role-based access control. The same audit trail.

A media company running WordPress VIP can connect any MCP-compatible agent to its 57 platform tools through one endpoint. Publishing, template updates, content migration — all governed, all logged, all with a kill switch.

A media company running Graphene-Selene can do none of this through protocol. Every AI integration requires custom engineering. Every external agent interaction requires bespoke API work. The proprietary advantage — deep knowledge graph, entity relationships, cross-brand content intelligence — stays locked inside the platform, visible only to internal tools.

The Salesforce Signal

The Contentful acquisition is the sharpest signal. Salesforce paid $1.5 billion — down from Contentful's $3 billion peak valuation — specifically to give Agentforce a structured content layer. Not for Contentful's customer base. Not for its revenue. For its content model. The architecture that lets agents compose experiences from structured content primitives.

Graphene-Selene has a richer content model than Contentful. It has knowledge graph relationships that Contentful does not. It has forty brands of content spanning food, health, beauty, entertainment, and lifestyle. It has the entity resolution and content intelligence that Salesforce is trying to build by bolting Contentful onto Agentforce.

But Contentful has a protocol surface. Graphene-Selene does not. The less capable platform is composable. The more capable platform is closed.

What This Means for October

I wrote in The Comment Period that October presents two deadlines for D/Cipher's ad tech layer. But October also presents the CMS question that nobody is tracking: as the entire CMS industry converges on MCP as the agent interaction standard, where does Graphene-Selene sit?

The intern postings in Edmonton show active development on Graphene-Selene services. The GitHub footprint shows no MCP surface. The D/Cipher PM job posting requires MCP expertise, but that posting is for the ad tech layer, not the content platform.

The question for People Inc's CTO is no longer "should D/Cipher have an MCP server." It is: should the company's two most valuable technology assets — its advertising intelligence and its content platform — both remain invisible to the agent ecosystem?

Adobe proved the architecture works with one URL. Seven vendors proved the market agrees. The CMS that powers 175 million monthly sessions across 40 brands has not shipped one.