Introducing the Airbyte MCP Server and Context Layer

Build pipelines, fix failing syncs, and query your data from Claude, ChatGPT, Cursor, or any MCP client.

DATA

October 27, 2026

Last updated October 27, 2026

Kushal Chatterjee

Summarize with AI:

You wanted it, you got it. Today, we’re officially releasing the Airbyte MCP server and context layer. The new MCP server was the most requested feature in Airbyte’s history, and we’re happy to finally launch it!

What exactly can you do with the Airbyte MCP server?

The Airbyte MCP server is open source, and free for Airbyte users on all plans, including Core. It lets Claude, ChatGPT, Cursor, VS Code, and any other MCP-compatible client run Airbyte for you. It can manage connections, diagnose failed syncs, manage custom connectors, and more!

For Airbyte customers on paid plans, we’ve added another cool feature. You can use the MCP server to access the new Airbyte context layer. It gives you the ability to chat with your data directly from Claude or ChatGPT. This means you get live records from supported sources, and direct reads from your Snowflake or BigQuery warehouse. The context layer is available to Airbyte Cloud and Enterprise Flex users, and can be enabled inside the Airbyte UI.

Let’s walk through the release and how it can support your workflows.

Run Airbyte from your agent

There’s a reason our users wanted this feature.

Everybody is working out of Claude, ChatGPT, and a plethora of other AI agents, and switching between UIs is an extra step that slows us down. This is as true for a product marketer like myself as it is for the engineers who use Airbyte. After all, why should you have to copy and paste a sync error message from the Airbyte UI into Claude to diagnose the error?

The Airbyte MCP removes that friction. In plain language, you can:

Build pipelines. Create sources, destinations, and connections, pick streams, set a schedule, and start a sync.
Fix failing syncs. Check sync status, read job logs, and get a solution, without digging through logs yourself.
Manage connectors. Review connector configurations and manage the custom connectors your team has built.
Get answers about Airbyte. The MCP searches Airbyte's docs, API specs, and GitHub, so you can ask how a connector behaves without leaving the conversation.
Chat with your data. Orgs on Airbyte’s paid plans get access to the context layer. It lets you chat with your Snowflake or BigQuery data warehouse inside your AI client, as well as query 50+ supported sources directly. More details in the next section.

A few prompts to start with:

"Were there any sync failures in the past 24 hours?"

"Why did my last Salesforce sync fail? Read the logs and suggest a fix."

"Create a connection from my Postgres source to my BigQuery destination that syncs every 6 hours."

The full list of capabilities is in the Airbyte MCP docs and the tools reference. Ian from our team also has a great demo video you can watch:

Chat with your tools and data warehouse directly

Outside of operational management, the MCP server gives you access to the Airbyte context layer (if you’re on one of our paid plans). The context layer can be enabled by your Airbyte admin inside the Airbyte UI.

Here’s why we think you should use it for your AI workflows.

There are two ways to get business context to an agent.

The first is runtime assembly. You hook your AI client to a bunch of MCP servers and let them ping the API every time the agent has to answer a question.

This approach is easy and intuitive inside Claude or ChatGPT, but you don’t get the organizational knowledge across multiple sources that you’ve accumulated and built over time. In essence, you lose the “gravity” of data, where data grows more powerful the more consolidated and centralized it is. You also burn tokens needlessly on search and pagination.

That’s why we believe you should point your agent towards your data warehouse, as that is the center of gravity for organizational context. This sounds simple in theory, but there are challenges.

As users, we know getting access to the BigQuery MCP server can be challenging. With the Airbyte MCP, you can easily connect your agent to your BigQuery or Snowflake from one endpoint. You now have direct, secure access to your entire organizational knowledge base.

And for sources that you don’t load into your data warehouse, the context layer supports direct reads to a growing set of source connectors, with more added every week.

The context layer docs cover setup and supported destinations.

Coming soon: semantic search

Direct reads and SQL work well when you know what you're looking for. A lot of business context lives in text instead: call transcripts, support tickets, issues, docs, and notes.

We are building semantic search into Airbyte connectors, because we feel it is vastly superior to keyword search for unstructured data. Keyword search requires you to know the names or metrics you are looking for, whereas semantic search actually queries data on meaning and lets agents query and act with less handholding.

Semantic search will be available for multiple Airbyte sources that are synced to BigQuery and Snowflake. Keep an eye out for further announcements!

Guardrails are built in

We take security for agentic workflows seriously. The Airbyte MCP server comes with several guardrails to make sure your AI agents access data with Airbyte safely.

Your permissions, nothing more. You sign in with your existing Airbyte account over OAuth or SSO, so there are no API keys to paste into a chat. The agent can only reach the workspaces and connections your role allows, and it can't elevate your access.
Confirmation before anything destructive. Every tool is classified as read, mutate, or destructive, and destructive actions like deletes require confirmation.
Read-only access to your data. Direct reads can list and get records, and warehouse queries are limited to SELECT statements.
Admins decide what agents can read. The context layer is off by default. An org admin turns it on, and workspace editors choose which sources and destinations agents can reach. Turning agent access off doesn't interrupt replication.
An audit trail. Actions taken through the MCP are logged, attributed to the user, tool, and target, and flagged in the Airbyte UI.

Where it works

Airbyte Cloud. The hosted MCP works with every Airbyte Cloud plan (Standard, Plus, Pro), and the context layer is available to every Cloud organization.
Airbyte Enterprise Flex. Enterprise Flex customers can use the hosted MCP too. The context layer runs on Airbyte's cloud infrastructure, so teams with strict sovereignty requirements may want to keep the context layer turned off.
Airbyte Core (open source). Run the MCP locally against your own instance. Pipeline management and troubleshooting tools work there, but you’ll need Airbyte Cloud to use the context layer. Please consult our guide to using the Airbyte MCP with Airbyte Core.
AI clients. Claude (web, Desktop, and Claude Code), ChatGPT, Cursor, VS Code, and any other MCP-compatible client.

Get started

Add the server. Point your client at https://mcp.airbyte.com/mcp. The install guide walks through each client.
Sign in. Your client opens a browser window so you can sign in with your Airbyte account.
(Optional) Turn on the context layer. If you're an org admin, enable agent access from the Context layer page in Airbyte.
Ask. Try "Which of my connections failed this week, and why?"

New to Airbyte? Start a 14-day free trial of Airbyte Cloud and connect any of our 700+ connectors.

Want to see it live? Join our walkthrough on October 28, 2026.

Found a bug or have a feature request? Open an issue on GitHub.

We’re excited to see how you use the Airbyte MCP server and context layer!