In the span of about two weeks, three very different software companies shipped the same thing.

  • GitLab made its MCP server generally available on October 15 with GitLab 19.5. It lets any MCP-compatible agent work across code, merge requests, work items and pipelines, and every action runs under the user's existing GitLab permissions. GitLab's docs cover setup for Claude Code, Claude Desktop, Cursor, GitHub Copilot, OpenAI Codex, Gemini and others. (GitLab)
  • Interact, an enterprise intranet vendor, announced Interact MCP on October 20: one permission-aware connection that gives approved assistants like Copilot, Claude, ChatGPT and Gemini 700 capabilities across 26 areas of the intranet. (Interact)
  • Traefik Labs launched the Traefik MCP Server on October 22, a read-only server that gives agents a gateway's valid schema, live configuration and observability data from Datadog, Grafana or Dash0. (Traefik Labs)

They weren't alone. Microsoft shipped the stable release of Azure MCP Server 3.0 on October 12. (Azure SDK Blog)

Is your SaaS agent-ready? Get your score

A developer platform, an intranet and an API gateway don't share buyers, markets or roadmaps. They made the same bet anyway.

The bet: your users are starting work somewhere else

Interact's announcement says it plainly: "Employees no longer start every task on the intranet. They ask Copilot, Claude, ChatGPT or Gemini, and more of them now have agents acting on their behalf."

Swap "intranet" for your product and the sentence still holds. More and more, the first stop for a task is an AI assistant, not your app. If the assistant can reach your product, the work flows to you. If it can't, the agent picks a tool it can use, and you never find out you lost the customer.

That's ghost churn: customers you lose without a cancellation, a support ticket or an exit survey, because an agent couldn't use your product and went with a competitor.

What these launches have in common

Read the three announcements side by side and the same design choices keep showing up:

  1. They work in the clients people already use. None of them ask users to adopt a new AI tool. They plug into Claude, ChatGPT, Cursor, Copilot and Gemini.
  2. They respect existing permissions. GitLab runs every action under the user's GitLab permissions, and Interact is permission-aware per person. Agents get exactly the access the human already has.
  3. They're designed for agents, not just wrapped around an API. GitLab says agents using raw APIs "often guess at endpoints and use up their context window on extra calls." Interact says its tools are "designed for AI agents rather than wrapped around a raw API."
  4. They go beyond search. These servers let agents act: create, edit, run workflows. Traefik is the one deliberate exception, staying read-only because of what it guards.

That's the bar now. A public API alone doesn't meet it.

Three questions to ask about your own product

  • Can an agent find you? Is there an llms.txt, clear API docs, or a published MCP server an assistant can discover?
  • Can an agent act, or only read? Agents that can only look things up won't move work into your product.
  • Does it run under your user's permissions? Without that, security reviews will stall agent access before it starts.

If the honest answer to any of these is "not yet," you're in the same spot most SaaS products are in today. The difference is that a growing list of companies, including the ones above, already have answers.

Get your product agent-ready

Pryno builds MCP servers and AI integrations so products like yours can be used by AI agents such as ChatGPT, Claude and Cursor. Think of it as SEO, but for AI.

Run the free Agent Readiness Check → or book a free AI-readiness call and we'll walk through where your product stands and what it would take to get agents using it.

Is your SaaS agent-ready? Get your score

Run a free Pryno readiness check on your MCP server