Context layer

Available today on Airbyte Cloud

Search and chat with your warehouse.

Your warehouse is your context. Give your AI tools data that's already synced and indexed. Admin control is built in.

Works with

  • Snowflake
  • BigQuery
The Airbyte context layer sits between your teams and your data. Business teams chat and search in Claude, ChatGPT, or Cursor. Data engineers trace a bad number to its pipeline. Your agents connect from any MCP client. It reads the data you sync to Snowflake and BigQuery and fetches live records from 50+ sources, read-only and admin controlled. Airbyte's 700+ connectors fill the warehouse.
  • 25,000+

    Companies run Airbyte

  • 1 in 5

    Fortune 500 companies

  • 2M

    Pipelines synced daily

  • 239%

    ROI, Certified by Forrester

Context that's ready before anyone asks.

Most AI tools assemble context at runtime, calling each app's API every time someone asks. The Airbyte context layer is pre-materialized: your pipelines already did the work.

Runtime assembly

A new round of API calls for every question

Raw, paginated payloads filling the context window

One app at a time

An MCP server to set up for every app

Pre-materialized context

Data already synced into Snowflake or BigQuery

Text already indexed for semantic search

Every source in one warehouse, ready to join with SQL

One Airbyte MCP endpoint for all of it

Ask the warehouse.
Search by meaning.
Fetch what's live.

Three ways your teams get answers, each behind a switch you control.

  1. 01

    Chat with your warehouse

    Ask questions of everything you sync to Snowflake or BigQuery. Answers inherit the organizational knowledge already in your warehouse.

    Warehouse

    > Overdue ARR, invoices past 30 days

     

    ● WAREHOUSE

    $1.84M · 14 accounts

     

    Northwind Bank · $412K

    Arrow Health · $268K

    Bluepeak Logistics · $231K

  2. 02

    Semantic search

    Search calls, tickets, and docs by meaning. It combines semantic and keyword matching, so it finds what keyword search alone misses.

    Semantic search

    > Data residency this quarter

     

    ● SEMANTIC SEARCH

    Northwind Bank · Negotiation

    "Stay in the EU."

     

    Arrow Health · Proposal

    "Our own AWS account."

  3. 03

    Real-time fetch

    Get live records that are fresher than your last sync, or from sources outside your warehouse. 50+ Airbyte connectors support real-time fetch.

    Real-time fetch

    > Northwind renewal still open?

     

    ● REAL-TIME FETCH

    Stage · Negotiation

    Close · This quarter

    Last touch · Today, 9:42 AM

    Priya Shah

When a number looks wrong, ask where it came from.

Airbyte loaded the data, so the Airbyte MCP server knows where every table came from. Ask Claude, ChatGPT, or Cursor why a dashboard looks off, and it checks the warehouse, finds the pipeline that loaded the table, and reads its last sync.

EXPLORE THE MCP

Airbyte MCP

> The revenue dashboard dropped 40% overnight. What happened?

 

● WAREHOUSE

SELECT max(_airbyte_extracted_at) FROM stripe.charges;

Last rows landed yesterday at 2:04 AM. Nothing since.

 

● PIPELINE

stripe.charges is loaded by the Stripe → BigQuery connection.

Its last sync failed at 3:00 AM: the Stripe API key expired.

 

Revenue isn't down. Yesterday's charges haven't synced. Update the Stripe credentials in Airbyte, then rerun the sync.

warehouse query: needs the context layer · pipeline check: every plan

You're in full control.

The Airbyte MCP connects AI tools to Airbyte. The context layer is what lets them read your data, and it stays off until you turn it on.

The context layer and semantic search are each off by default. Only an organization admin can enable them, as a one-time switch for the whole organization.

The context layer and semantic search are each off by default. Only an organization admin can enable them, as a one-time switch for the whole organization.

Context layer settings

Context layer off. Semantic search off. Only organization admins can change this.

  • Read-only

    SQL is limited to SELECT and SHOW TABLES, and real-time fetch only lists and gets records. Agents can't change data through the context layer.

  • Your permissions

    The agent acts as the signed-in person and can't raise their access.

  • Credentials stay in Airbyte

    Queries use the credentials saved on the connector. The agent never sees them.

  • Logged

    Every action is recorded with the user, tool, and target.

Where it works, and where your data goes.

Need data to stay inside your boundary? Run replication on Airbyte Flex and talk to us about the context layer.

Works with Snowflake and BigQuery. Use it from Claude, ChatGPT, Cursor, or any MCP client. Included on Airbyte Cloud Standard, Plus, and Pro, and Enterprise Flex. Not included on Airbyte Core. The 14-day trial on Standard and Plus includes the context layer and semantic search.

  • Warehouses

    SnowflakeBigQuery
  • AI tools

    ClaudeChatGPTCursor+ Any MCP client
  • Plans

    Standard14-day trialPlus14-day trialProEnterprise FlexCore

No. The context layer and semantic search are each off until an organization admin turns them on. After that, you choose which sources and destinations agents can read, and you turn on semantic search connection by connection.

The Airbyte MCP connects your AI tool to Airbyte, and it's free on every plan, including Airbyte Core. On its own, it manages pipelines and answers questions about Airbyte. The context layer is what lets it read your data: chat with your warehouse, semantic search, and real-time fetch, on Airbyte Cloud Standard, Plus, and Pro, and Enterprise Flex.

No. The context layer is read-only. Warehouse queries are limited to SELECT and SHOW TABLES, and real-time fetch can only list and get records.

People sign in with their own Airbyte account, and the agent gets their Airbyte permissions and can't raise them. Warehouse queries run with the credentials saved on the connector, not each person's own warehouse login, so give that connector a read-only role that covers only what agents should see. Every action is logged with the user, tool, and target.

Chat with your warehouse runs as SQL in your own Snowflake or BigQuery, on your compute. Semantic search runs on Airbyte Cloud or Airbyte-managed data planes, and Airbyte runs the embedding model on its own infrastructure, so your text isn't sent to a third-party embedding API. Real-time fetch runs on Airbyte Cloud. Answers go to the AI tool you connect. If your data has to stay inside your own boundary, run replication on Airbyte Flex and talk to us about the context layer.

Keyword search only finds the exact words you type. Semantic search finds what people meant, so "I got charged twice" turns up in a search for billing complaints. Airbyte combines semantic and keyword matching, so exact order numbers, product names, and IDs still match too.