Editorial note · MCP & Integration · 4 min read
Why MCP changed our DAM integration roadmap.
A short field note on the protocol that quietly reorganized the AI-meets-DAM stack in 2026 — and the reason most vendors are now a year behind.
For the first time in a decade, the way creative teams will pull assets out of their DAM is going to change. It's not because of a new vendor, a new model, or a new format. It's because of a protocol.
The Model Context Protocol — MCP — is a standard for how language models call tools. Anthropic shipped the first version in late 2024. It quietly became the dominant convention through 2025. By early 2026, every major LLM client — Claude Desktop, ChatGPT Desktop, Gemini's coding mode, every IDE that integrates AI — speaks MCP natively.
What MCP does for DAM teams is small in theory and huge in practice: it removes the integration cost between "asset library" and "language model." Drop three lines into a config file. Restart your client. Your library is now queryable.
Before MCP, every DAM-to-LLM integration was bespoke. You wrote tool definitions in the SDK. You proxied auth. You mapped schema. You handled pagination. You debugged JSON shape drift on the first real query. The honest median, which we measured in Report 01, was about an hour for a competent engineer. For non-engineers, it was a never.
After MCP, an operator at a creative team installs the same integration in under two minutes. The same integration. No code.
What broke (and why most vendor roadmaps are wrong)
Every DAM vendor we surveyed has "AI" on their 2026 roadmap. Almost none of them have MCP as the central architectural pattern. Instead, they have:
- "AI tagging" features bolted onto the upload pipeline.
- "AI search" features bolted onto the UI.
- "AI copilot" features bolted onto the dashboard.
All of these are useful. None of them are the thing. The thing is that the operator now expects to query the DAM from inside the LLM they're already using, not from inside the DAM's UI. The integration surface is the LLM, not the DAM. If your DAM doesn't ship an MCP server, you're not on the integration surface.
This is why 1 of 6 vendors shipping native MCP is a category-defining gap. Through H2 2026 the rest will follow — we'll re-verify that statistic weekly — but until they do, the buying decision is much simpler than the AI-feature checkboxes suggest.
What this changed for our research
When we started DAM LLM, the plan was a deep series on AI tagging accuracy — precision and recall benchmarks across providers. That's still coming (Report 03). But we reordered the queue. Report 01 — the install-time field study — is now the first thing readers see because it's what changed most under MCP.
If your team is in a 2026 DAM evaluation right now and the vendor's AI roadmap doesn't mention MCP by name, ask why. There's no good answer. There's only "we're working on it" — and the question worth asking back is, on what timeline.
— Itai