Use CIN MCP in Agent Builder

Register CIN MCP as an MCP server in Agent Builder and give an agent some of its tools, so the agent can call Content Intelligence itself.

Prerequisites

An Agent Builder role that can create agents and register MCP servers (cin-agent-builder.solution-builder or .admin).
An AI client connected to CIN MCP (see Tutorials), or the Agent Builder API.

Tutorial Overview

Register the CIN MCP server.
Create a tool agent that uses it.
Invoke the agent.

How authentication works here

Agent Builder requires an API key when you register an MCP server. CIN MCP doesn't use API keys: it acts on the invoking user's Hyland token, which Agent Builder sends with each tool call. So the registration takes a placeholder key, and each tool call runs with the permissions of whoever invokes the agent.

note

The placeholder key protects nothing and grants nothing. Access is controlled entirely by the invoking user's Hyland roles.

Part 1: Register the server

Ask your AI client:

Register an Agent Builder MCP server named "cin-mcp" at https://cin-mcp.ai.app.hyland.com/mcp, private, validated immediately, with a placeholder API key.

This calls agent_builder_register_mcp_server with:

Argument	Value
`name`	`cin-mcp`
`url`	`https://cin-mcp.ai.app.hyland.com/mcp`
`validation_mode`	`immediate`
`visibility`	`private`
`api_key`	Any placeholder, for example `placeholder-not-used-by-cin-mcp`

Expected result: the registration with "status": "active", validation completed and health healthy. Note its id.

Check the tools Agent Builder can see:

List the tools of MCP server [id].

(agent_builder_list_mcp_server_tools; it returns the CIN MCP tool list.)

Part 2: Create a tool agent

Give the agent only the tools it needs. For a first test, three read-only tools:

Create a tool agent named "cin-mcp-access-test", temperature 0, that uses MCP server [id] with only agent_builder_list_models, agent_builder_list_agent_types and content_lake_list_graph_databases.

This calls agent_builder_create_agent with agent_type tool and a tool entry:

{
  "toolType": "mcp",
  "name": "cin_mcp",
  "description": "Hyland Content Intelligence MCP.",
  "mcpServerReferences": [
    {
      "id": "<registration id>",
      "validationMode": "lenient",
      "toolConfiguration": {
        "allowedTools": [
          "agent_builder_list_models",
          "agent_builder_list_agent_types",
          "content_lake_list_graph_databases"
        ]
      }
    }
  ]
}

Part 3: Invoke the agent

Invoke agent [id], latest version: "Call agent_builder_list_models, agent_builder_list_agent_types and content_lake_list_graph_databases (limit 3), and say whether each succeeded."

The agent calls the three tools and reports the available models, the agent types, and three of your environment's graph databases.

Best Practices

Least privilege. List only the tools the agent needs in allowedTools. Prefer read-only tools; add tools that change data only with human-in-the-loop approval.
Private registration. Keep the registration private unless the whole environment should use it.
Clean up. Delete test agents and registrations (agent_builder_delete_agent, agent_builder_delete_mcp_server) when you're done.

Summary

Register CIN MCP in Agent Builder with a placeholder API key, attach it to a tool agent with an allowedTools list, and invoke the agent. The agent's tool calls carry the invoking user's permissions.

Prerequisites​

Tutorial Overview​

How authentication works here​

Part 1: Register the server​

Part 2: Create a tool agent​

Part 3: Invoke the agent​

Best Practices​

Summary​