A connector is a prebuilt piece of software that links one system to another so that data, and increasingly actions, can pass between them without anyone writing integration code. The person using it authorises an account, picks a few settings and receives a working link. The code that authenticates, requests, translates and returns the data has already been written by someone else.

Connectors exist because marketing data is scattered. Campaign results sit in ad platforms, revenue sits in a customer relationship management (CRM) system or a commerce platform, and analysis happens somewhere else. Each pairing could be wired by hand through an application programming interface (API), but that means authentication, schema mapping and maintenance for every combination. A connector packages that work once and sells or gives it to everyone who needs the same link.

How a connector works

The word covers three families in advertising, and the mechanics differ in each.

Reporting connectors pull data into dashboards. Google's Data Studio, the reporting tool that carried the Looker Studio name until April 2026, is the reference case. Its community connectors are Apps Script projects that must define a fixed set of functions, according to Google's developer documentation. The getAuthType function declares how the connector signs in to the third-party service, with options including no authentication, a key, a username and password, or OAuth 2.0, the delegated authorisation standard. The getConfig function describes the settings screen the user fills in, including whether a date range is required. The getSchema function lists the available fields, typed as dimensions or metrics, and getData fetches the rows. Each request carries a lastRefresh timestamp so a connector can decide whether to serve cached results or call the source again, and Google's API reference states that a request covering more than 20 fields is split into multiple data calls. A manifest file, appsscript.json, holds the name, description and listing details.

Data ingestion connectors push first-party data into ad platforms for targeting, measurement and bidding. In Google's Data Manager, a connection links a source such as BigQuery, Google Drive, HubSpot or Shopify to Google's advertising products. The documented import windows show how much configuration hides behind the label. For Google Cloud Storage, Amazon S3, HTTP, SFTP and Google Sheets, each run reaches back 90 days for conversions. Salesforce and HubSpot import 14 days on the first run and only changed records afterwards, while BigQuery, Amazon Redshift, Snowflake, MySQL and PostgreSQL import the last 14 days on every run.

AI connectors link a conversational assistant to outside software. In Anthropic's Claude and, until a rename, OpenAI's ChatGPT, a connector is the product wrapper around a server that speaks the Model Context Protocol (MCP), the open standard for how an AI application discovers and calls external tools. The server exposes tools the model can invoke, resources it can read and prompts a person can select. According to Anthropic's help centre, custom connectors are reached from Anthropic's cloud infrastructure rather than the user's device, so the server must be reachable over the public internet, and free-plan users are limited to one custom connector. Adding a custom connector takes a name, the server URL and, optionally, an OAuth client ID and secret. Anthropic's developer guidance lists support for both Server-Sent Events and Streamable HTTP transports, warns that the former may be deprecated, and confirms support for Dynamic Client Registration, in which the assistant registers itself with the server's authorisation system automatically. A directory connector has passed Anthropic's review; a custom one has not.

The economics resemble those of reporting connectors. Wiring a dozen AI applications to a dozen data sources by hand requires 144 separate integrations; a shared protocol reduces that to 24, because each side implements the standard once.

Operators sit on both sides. Platforms build first-party connectors for their own data. Specialist vendors such as Supermetrics, Windsor.ai and Dataslayer build connectors to other companies' platforms, many of them sold on subscription. Agencies build internal ones for their own accounts.

Origin and evolution

Google announced Data Studio on March 15, 2016 as part of the Google Analytics 360 Suite, and a free version followed in May 2016. The product initially shipped with connectors for Google's own services and a handful of general sources such as file upload and MySQL. Google launched community connectors in late 2017, opening the tool to third-party developers. By November 2017, more than 600 developers had signed up for access, Google's analytics blog reported, as it opened a repository for open-source connectors. Google's review of 2018 counted more than 140 partner connectors and over 500 data sources.

AI connectors followed the protocol. Anthropic published MCP in November 2024. On May 1, 2025, Anthropic launched Integrations, linking Claude to remote MCP servers with 10 initial services including Atlassian, Zapier and Cloudflare. A curated directory of connectors followed on July 14, 2025, and HubSpot released the first CRM connector for Claude on July 29, 2025.

Advertising platforms then built their own endpoints. Google released an open-source, read-only MCP server for the Google Ads API on October 7, 2025, and Amazon Ads opened a closed beta for its server on November 13, 2025, supporting Claude, ChatGPT, Amazon Q and Amazon Bedrock. On April 29, 2026, Meta opened its ad system through connectors built on its ads MCP server and a companion command line interface, with write access to campaigns from the first day.

The vocabulary moved in parallel. OpenAI's help centre states that on December 17, 2025, connectors were renamed apps, a term now covering both interactive apps and integrations that search a user's information. On July 9, 2026, OpenAI migrated its app directory into a plugin directory, where a plugin can bundle skills, apps and app templates.

Why it matters for marketers

Connectors decide which numbers reach a report without manual exports. Data Studio supports more than 600 partner connectors, and its Google Ads and Google Analytics connectors are free, while many third-party connectors require paid subscriptions. Field coverage changes over time as well: Google added seven fields to the Google Ads connector on October 2, 2025, including conversions by conversion date and gross profit margin.

On the activation side, connectors carry the first-party signal that automated bidding consumes. Google has published customer results for Data Manager, including Multiplier Technologies recording 25% more qualified leads after connecting HubSpot conversions. The figure is self-reported through Google and has not been independently verified.

AI connectors change who can operate an account. Anthropic stated in June 2026 that its directory had grown to more than 200 connectors since July 2025. When a connector carries write permissions, the assistant is no longer only reading a report but changing budgets and campaigns.

Limitations and disputes

A connector is only as stable as the API beneath it. PPC Land's analysis of Google's December 2025 connector release noted that partner connector reliability depends on upstream API stability, and that several vendors often sell connectors to the same source. Upstream policy also constrains output: the BigQuery Data Transfer Service connectors for Google Ads and Search Ads 360 stopped backfilling beyond 37 months under a retention change dated June 1, 2026. Where no connector exists, CSV upload remains the workaround.

Security is the sharper dispute. Machine learning engineer Akshay Pachaar wrote on July 20, 2025 that "MCP security is completely broken", pointing to tool poisoning attacks that inject instructions into what a model treats as data. Anthropic warns that custom connectors reach services it has not verified.

Platforms disagree on how much intermediation to allow. Google's developer policies, updated on August 31, 2026, prohibit programmatic proxies, which Google defines to include wrapper services and MCP servers that solely re-expose Google Ads capabilities, citing cross-tenant data leaks and throttling. Hosted multi-client connectors built on shared credentials fall inside that definition. Meta took the opposite direction on July 16, 2026, opening its ads MCP server to any developer holding a Meta app.

Not the same as

An API is the interface a platform publishes. A connector is prepackaged software that uses one on a user's behalf.

An MCP server is the endpoint exposing tools. The connector is the client-side listing and authorisation that attaches that server to an assistant.

A plugin, in OpenAI's 2026 usage, is a bundle that may contain an app, skills and templates, not a single link. OpenAI's original plugin store, shut down in April 2024, was a different product.

A pixel or tag collects events in a browser. A connector moves data between systems after collection.

Recent developments

Semrush published an official Claude connector on August 26, 2026, describing it as its third MCP integration with a major model without naming the other two. X launched an ads server on August 24, 2026 exposing 23 tools, ten of which write to live accounts. On September 14, 2026, Snipp Interactive released a Claude connector and a ChatGPT plugin for brand rebates and loyalty offers, a sign that connectors are moving from reporting into commerce. As of September 2026, Google's compliance team is reviewing existing Ads API integrations against the proxy rules.

Timeline

  • March 15, 2016 - Google announces Data Studio as part of the Google Analytics 360 Suite
  • May 2016 - A free version of Data Studio becomes available
  • Late 2017 - Google launches Data Studio community connectors for third-party developers
  • November 2017 - Google opens an open-source community connector repository, reporting more than 600 registered developers
  • 2018 - Google counts more than 140 partner connectors and over 500 data sources
  • November 2024 - Anthropic publishes the Model Context Protocol
  • May 1, 2025 - Anthropic launches Integrations, connecting Claude to remote MCP servers
  • July 14, 2025 - Anthropic launches its connectors directory for Claude
  • July 20, 2025 - Tool poisoning weaknesses in MCP implementations are publicised
  • July 29, 2025 - HubSpot releases the first CRM connector for Claude
  • October 2, 2025 - Google adds seven fields to the Google Ads connector in Looker Studio
  • October 7, 2025 - Google releases a read-only open-source MCP server for the Google Ads API
  • November 13, 2025 - Amazon Ads opens a closed beta of its MCP server
  • December 17, 2025 - OpenAI renames ChatGPT connectors to apps
  • December 23, 2025 - Looker Studio adds 15 partner connectors
  • April 11, 2026 - Google announces the return of the Data Studio name
  • April 29, 2026 - Meta launches Ads AI Connectors with write access
  • June 1, 2026 - 37-month limit applies to Google Ads granular reporting data
  • July 9, 2026 - OpenAI migrates its app directory to a plugin directory
  • July 16, 2026 - Meta opens its ads MCP server to any developer with a Meta app
  • August 24, 2026 - X launches an ads MCP server with 23 tools
  • August 26, 2026 - Semrush publishes its official Claude connector
  • August 31, 2026 - Google prohibits programmatic proxies, naming MCP servers
  • September 14, 2026 - Snipp Interactive releases a Claude connector and a ChatGPT plugin

Summary

Who. Platforms such as Google, Meta, Amazon Ads and X build connectors and servers for their own systems. Specialist vendors including Supermetrics, Windsor.ai and Dataslayer sell connectors to other companies' platforms. Anthropic and OpenAI run the directories through which AI connectors are distributed, and agencies build internal connectors for their own accounts.

What. A connector is prebuilt integration software that authenticates to one system and moves data or actions into another. In advertising it takes three main forms: reporting connectors feeding dashboards, ingestion connectors sending first-party data to ad platforms, and AI connectors attaching MCP servers to assistants.

When. Reporting connectors became mainstream after Google launched Data Studio in 2016 and opened community connectors in late 2017. AI connectors date from Anthropic's Integrations launch on May 1, 2025, with advertising platforms following between October 2025 and August 2026.

Where. Connectors run inside reporting tools such as Data Studio, inside ad platform data hubs such as Google's Data Manager, and inside AI assistants such as Claude, where remote connections originate from the provider's cloud rather than the user's device.

Why. Marketing data is spread across dozens of systems, and hand-built API integrations are costly to write and maintain. Connectors trade that cost for dependence on a vendor, on upstream API stability and, for AI connectors, on security and platform policies that are still being contested.