Merge Docs

Merge Agent Handler

Merge Agent Handler securely connects AI agents to third-party tools via the Model Context Protocol (MCP). It manages authentication, permissions, and monitoring of all tool interactions.

Agent Handler includes:

Prebuilt connectors for common enterprise systems
Built-in authentication and credential handling
Observability and data-loss prevention controls

Key concepts

The high level flow of Merge Agent Handler is: you register a user so they can connect to third party data providers. That user then takes actions through your agents. As users take actions, Merge Agent Handler scans tool inputs and responses for sensitive data, and prevents data misuse based on configurations you set in the platform.

Concept	Description
Registered User	The users who will be interacting with your agent. You can create these programmatically.
Connectors	The third-party systems that your customers will interact with. Examples: Jira, Linear, and Asana.
Tools	The specific actions within a connector that your customers can take. Examples: Create_Issue tool in Jira, a Get_Issue tool in Linear, or a Create_Task tool in Asana.
Tool Pack	The bundling of connectors and specific tools that your customers can use through your agent.
Link	UI component that guides your app’s user through authentication for a connector.
Connector Studio	The suite to add or customize connectors and tools for your different use cases.
Security Gateway	The security gateway scans tool inputs and responses for sensitive data, and takes action based on what’s configured within the platform.

Data flow diagram

How your product will interact with Merge Agent Handler

Merge Agent Handler sits in-between your AI agent and third-party tools.

A workflow could look like:

Your user will call Merge Agent Handler to list all tools available for your registered user to use
Merge Agent Handler will respond with the available tools enabled in the Tool Pack
Your registered user prompts your agent to connect. Your agent calls one of the available tools. If it’s your user’s first time interacting with a given connector, we’ll signal to prompt for the user’s credentials via Link.
The tool call arguments are passed through the Security Gateway & DLP solution to scan for sensitive entities.
The tool call is executed and the requests are made to the third-party.
The responses are returned by the third party. They will pass through the Security Gateway once more, and then returned back to your agent.