MCP Connector.
An MCP connector (or simply connector) links an AI assistant such as Claude or ChatGPT to another service, for example a CRM, a calendar or a file store. It gives the assistant tools such as 'find contact' or 'add appointment' and usually runs on the Model Context Protocol (MCP).
MCP Connector — Explained in Detail
A language model knows nothing about your customers, appointments or documents by itself. A connector fills that gap. It is a small program, known as an MCP server, that offers an AI assistant a list of tools: read, search, create, update. During a conversation the assistant decides whether it needs a tool, calls it and carries on with the result. The underlying principle is tool calling.
The connector is not the same as the protocol. The Model Context Protocol is the open standard Anthropic introduced in November 2024, now supported by many vendors. A connector is a specific integration built on that standard, for HubSpot or Google Drive, say. Claude calls these integrations connectors; in ChatGPT the equivalent entries are ChatGPT apps, which since a reorganisation sit in a directory called 'Plugins'. Either way, an assistant gains the ability to act through someone else's interface.
We counted the directories ourselves on 11 November 2026, signed out and from Switzerland: 1,111 entries in Claude's directory and at least 6,126 in the anonymous view of ChatGPT's. What matters more to a small business is whether its own tools are among them. In the same count, the Swiss accounting software bexio was missing from both directories.
A connector gives an AI system access to real data and sometimes the right to change it, so permissions come first. According to the Claude Help Center ('Use connectors', retrieved 11.11.2026), each tool of a connector can be set to always allow, needs approval or blocked; the OpenAI Help Center ('Connected apps in ChatGPT', retrieved 11.11.2026) describes levels from read-only to all actions. A sensible start is read access only, confirmation for anything that writes, and caution with content from external sources, see prompt injection and guardrails.
Ready-made connector or your own integration
For widely used services such as Microsoft 365, Google Drive, HubSpot or Slack, vendors publish ready-made connectors. They take minutes to switch on but only cover the tools the vendor chose to expose. If a service is missing, such as industry software or your accounting package, you need your own MCP server. It talks to the service's API and defines what the assistant may see and which actions it may take.
We build custom integrations as part of AI agent development. Which processes justify one, and where simple automation is enough, we work out through AI automation for SMEs or AI consulting.
Example: a property manager connects Claude to its CRM
A constructed example: a property management firm in St. Gallen uses a CRM for which a ready-made connector exists. Staff want to ask the assistant questions like 'Which applicants for the flat on Rosenbergstrasse have we not contacted yet?' The firm enables the connector in its company account and initially allows only the read tools.
After a few weeks it unlocks 'add note', with confirmation before every use. Deleting contacts stays blocked. For the property accounting system, which has no connector, it evaluates a custom MCP server. The assistant saves search time while the team keeps control over changes, see human-in-the-loop.
Related Page
AI agent developmentFrequently Asked Questions About MCP Connector
MCP is the open protocol, the shared language between AI assistant and service. A connector is a specific integration that uses the protocol, for a CRM or a calendar for example.
With the major assistants, yes. Claude lets you set each tool to always allow, needs approval or blocked; ChatGPT offers levels from read-only to all actions. Starting with read-only tools is a sensible default.
If the software has an API, a custom MCP server can be built. It defines which data the assistant sees and which actions it may perform. Without an API, only workarounds remain, such as exports or computer use.
More Terms Starting with "M"
Meta Description
The meta description is an HTML tag that contains a brief summary of a webpage for search engines. I...
Mobile-First Index
Since 2019, Google primarily crawls and indexes the mobile version of a website (no longer the deskt...
Multimodal Search
Multimodal search lets people search with several input types at once, such as text, image, voice or...
Ready for Your Project?
Apply this knowledge to your website — DLM Digital will help you.