MCP connectors: how they work and real-world examples

Model Context Protocol (MCP) connectors are often mis-categorized, leading to confusion when teams try to use the protocol with their agents.

To help clarify what an MCP connector means, we’ll align on a shared definition, walk through how you can use them, and break down real-world examples.

What is an MCP connector?

An MCP connector is an MCP server that’s either provided by the software solution or built and maintained by a 3rd-party. 

Every MCP connector consists of exposable tools that allow agents to access certain types of data and perform specific tasks.

Related: What is MCP observability?

How MCP connectors work

Here’s a general workflow:

1. A user inputs a prompt to your AI agent.

2. The agent identifies the appropriate MCP connector and the right tool within that server.

3. The agent would ask the user to authenticate with the MCP connector to confirm they have the right set of permissions.

4. After successful authentication, the agent invokes the tool and passes along the appropriate arguments to perform the requested action.

5. Once the operation is completed, the agent confirms to the user that the action was successfully performed and, if applicable, provides links or references for verification.

To make this more tangible, here’s how this can work if you’re using a GitHub MCP connector: A user can ask your agent to create an issue in GitHub that’s high priority, needs to be resolved by the end of the week, should be assigned to Jane Doe, and includes a certain issue title and description.

The agent reviews the user’s existing permissions with GitHub, and if they don’t have permissions to create an issue, they’ll prompt the user to authenticate. 

Once authenticated, the agent will call the GitHub MCP connector’s create_issue tool with parsed arguments like title, body/description, assignees, labels, and the repository. 

Once the issue is created, the agent confirms that it was successful to the user and returns the URL to the new issue.

Real-world examples of MCP connectors

The quality of MCP connectors for a single software application can vary widely, as the MCP connector providers (i.e., the software application and 3rd-party providers) can offer varying tools, tool names, and tool descriptions.

That said, we’ll walk through how a few MCP connectors can work based on what’s supported by Merge Agent Handler.

Linear

Linear helps product teams manage issues, projects, and product roadmaps.

Its MCP connector supports OAuth 2.0 and allows users to retrieve, create, update, and resolve comments and issues; manage projects; and retrieve or update users and teams.

As an example, Perplexity offers a Linear MCP connector to help users get context on issues or projects and combine that information with other data sources to uncover broader insights. The connector also lets users create or update issues without leaving Perplexity. 

Related: Examples of how you can use MCP

Salesforce

You can use an MCP connector for the CRM platform to search, get, create, update, delete, and list accounts, cases, contacts, notes, leads, and more. And like Linear, the Salesforce MCP connector can also be authenticated via OAuth 2.0.

Slack

The MCP connector for the business communications platform can be authenticated via OAuth 2.0.

It can allow users to perform a wide range of actions on channels, such as joining, creating, or leaving any. You can also post, update, schedule, pin, and delete messages; and even get and add details on users, like their statuses, conversations, and profile information.

Benefits of using MCP connectors

The benefits of using MCP connectors largely depends on the provider. That said, here are some potential benefits from using them.

• Third-party solutions offer pre-built and fully-maintained connectors, saving your engineers countless time 

For example, Merge Agent Handler offers 30+ pre-built connectors with applications across software categories, enabling your engineers to avoid connecting to any and maintaining the connections.

• If you can’t find an existing MCP connector, you can use a 3rd-party solution to auto-generate them 

For example, Merge Agent Handler lets you simply provide the connector name and a few sentences that cover the requirements and specifications you need the connector to meet. From there, Merge can use AI to generate it and the team will go on to review it before sharing it with you.

‍

• MCP connectors can be easily customized, whether that’s editing the tool names, descriptions, and schema 

This helps you personalize the tool for your specific agentic workflows and allows you to make tweaks to optimize evaluation results during testing.

• You can enforce authentication for each connector to prevent data leaks and follow the principle of data minimization

• You can test MCP connectors’ tools in a variety of ways before moving them to production

For example, for a given connector, Merge Agent Handler lets you set custom prompts, define the desired behaviors (i.e., tool calls and output formats), and see whether your agent behaves as expected

{{this-blog-only-cta}}