Airbyte Agent MCP

Hosted MCP server from Airbyte: AI assistants reach connected services such as applications and data sources via OAuth, with nothing to install.

Description

The Airbyte Agent MCP connects AI assistants such as Claude Code, Cursor, Codex, or ChatGPT to the services where your data actually lives. Airbyte is best known as a data integration provider, that is, for moving data from applications into data warehouses. With Airbyte Agents the company extends this approach: an assistant gets authenticated access to connected platforms and can list, search, and retrieve records there, and in some cases create or change them. Airbyte runs the server itself, so according to the documentation you do not need to install anything. It is reachable at https://mcp.airbyte.ai/mcp and uses Streamable HTTP as transport.

What can the Airbyte Agent MCP do?

According to the documentation the server works with connectors: each service you connect gets its own connection with its own access. The assistant can add and remove connectors and query the data behind them. The page names tools around describing the connectors: inspect_connector, read_skill_docs, list_skills, and search_skills. With them the assistant learns at runtime what a connector can do instead of holding all functions in context up front. The page does not give a complete tool list; we therefore claimed no more than it says.

Two layers of sign-in

According to the provider, sign-in runs in two steps. First your client starts an OAuth 2.0 flow, you log in to your Airbyte account in the browser, and the client stores the token. After that every connector has its own flow, either through an OAuth consent screen or a credential form. The documentation stresses that credentials are always entered in the browser and never in the chat. That is a sensible safeguard, because passwords and keys do not end up in the assistant's conversation history.

Setup in the common clients

In Claude Code one command is enough, and you then confirm the sign-in via /mcp. For Claude Desktop you add a custom connector with the server address under Connectors in the settings; the documentation explicitly warns against using the file claude_desktop_config.json for this. Codex, Cursor, and VS Code are described as well, as is a ChatGPT entry called Airbyte Agent Engine, which according to the page is currently available only in the United States. Any client that supports OAuth and Streamable HTTP can use the address.

What is it particularly useful for?

The value lies where questions touch several systems: an assistant can look up tickets, customer data, or project data from different applications without you setting up a separate server for each service. Compared with a single-service server such as those from Fivetran or dbt, the focus here is not on managing pipelines but on direct access to the source data. Note that Airbyte also lists other MCP servers, including a knowledge server for the documentation and a server for Airbyte Cloud that the platform documentation describes as being in a private beta; these are not part of this entry.

Security and privacy

With connectors the assistant gains access to real account and business data. Connect only the services you really need, and check the requested permissions on every consent page. Where write access is possible, a mistake by the assistant can change records. Retrieved content enters the context of your AI client and therefore reaches the model provider; content from connected services can also contain instructions the assistant should not follow.

Costs

According to the documentation the feature is available in the Free, Individual, Team, and Custom plans; the exact terms are on the official Airbyte pages. The pricing page describes paid cloud plans and free self-hosting of Airbyte Core, but gives no separate breakdown for the agent plans. Check the current terms before use.

Who is the Airbyte Agent MCP for?

It is worthwhile for teams that want assistants to work on data from several SaaS applications and can use an Airbyte Agents account for it. Start with a single, uncritical connector and read-only queries, and expand only once you know the permissions and behavior. If you only want to connect one system, the dedicated server from that provider is often simpler.

Requirements

An Airbyte Agents account (app.airbyte.ai), an MCP client that supports OAuth and Streamable HTTP, and credentials for every service you want to connect.

Installation instructions

Register the hosted server https://mcp.airbyte.ai/mcp via HTTP in the MCP client, sign in to Airbyte in the browser, and then connect the desired connectors through their own credential flow.

claude mcp add --transport http airbyte-agent https://mcp.airbyte.ai/mcp

Authentication

OAuth 2.0 with the Airbyte account; in addition each connector has its own credential flow in the browser, either via OAuth consent or a form.

Required access permissions

Access to the connected services within the scope of the granted permissions: listing, searching, and retrieving records, in some cases also creating or changing them, and managing the connectors.

Transmitted or stored data

Retrieved records from connected services enter the AI client context and therefore reach the model provider. Processing by Airbyte follows the provider's privacy policy.

Security risks

Access to real business data, unintended changes with write permissions, exposure to the model provider, and prompt injection via content from connected services.

License and costs

License
Not recorded yet.
Cost
free

According to the documentation available in the Free, Individual, Team, and Custom plans; details are on the official Airbyte pages. For Airbyte Core itself the pricing page mentions free self-hosting.

Alternatives

Not recorded yet.

At a glance

Provider
Airbyte
Status
Official server
Deployment
Remote
Current version
Not recorded yet.
Last reviewed
23.10.2026

Repository and documentation

Categories

Supported clients