On October 27, Airbyte officially released its MCP server, and the launch post opens with a line most SaaS founders should read twice: the new MCP server "was the most requested feature in Airbyte's history." (Airbyte)
Airbyte has more than 700 connectors and a large open-source community. Out of everything that community could have asked for, the top request was a way to run Airbyte from Claude, ChatGPT, Cursor or VS Code.
What Airbyte actually shipped
The Airbyte MCP server is open source and free on every plan, including Core. With it, an AI agent can:
- build pipelines by creating sources, destinations and connections, then starting a sync,
- fix failing syncs by reading job logs and explaining what went wrong,
- manage the custom connectors a team has built, and
- answer questions about Airbyte from its docs, API specs and GitHub.
Paid customers also get a "context layer" that lets them chat with live data from supported sources and query Snowflake or BigQuery directly from their AI client. (Airbyte)
Airbyte also didn't skip the safety work. Its tools are labelled read, mutate or destructive, destructive actions need explicit confirmation, and every action is logged with the user and tool that made it. (Airbyte MCP)
It isn't just Airbyte
Earlier this month Microsoft shipped Azure MCP Server 3.0 as a stable release (Azure SDK Blog). In the same few weeks GitLab, Traefik and Interact all launched MCP servers too, which we covered in our roundup.
The pattern is consistent. Customers are doing more of their work through AI agents, and they want the tools they already pay for to work inside that agent.
The quiet risk: ghost churn
Most customers won't file a feature request. They'll start running a workflow through an agent, find that your product isn't reachable, and do that part of the job in a competitor's tool that is. Usage drops slowly and nothing in your dashboards explains why. That's ghost churn.
Airbyte heard the demand loudly because it has a large, vocal community. Most SaaS products won't get that warning.
What to do this quarter
- Find out what agents can see today. Check whether your product publishes an
llms.txt, an OpenAPI spec, public API docs and an MCP server that agents can discover. - Pick the three to five jobs customers repeat most. Those become your first MCP tools.
- Label risk the way Airbyte did. Separate read, write and destructive actions, and ask for confirmation on anything destructive.
See where your product stands
Our free Agent Readiness Check scans your site for the signals AI agents look for and scores you in about a minute.
