Integrations
============

The **Integrations** tab is where Super Administrators manage the tools and MCP servers available to agents across all workspaces. It contains two sub-tabs: **In-House Tools** and **MCPs**.

Disabling a tool or MCP at the platform level removes it from every workspace immediately — workspace administrators cannot re-enable them.

In-House Tools
--------------

In-House Tools are built-in integrations that agents can call during execution — search APIs, calculators, code runners, and similar utilities.

.. image:: /images/super-tools.png
   :alt: In-House Tools
   :align: center
   :width: 70%

|

Each tool card shows:

- **Tool Name** — the name of the tool (e.g., Calculator, Exa Search, Google Serper API)
- **Description** — what the tool does
- **Configuration Status** — whether additional setup is required before the tool can be used, or whether it works out of the box
- **Available in** — which agent types support the tool (Agentic Flow, Voice Agent)
- **Enabled toggle** — current availability across all workspaces

Use the toggle to enable or disable a tool. Common reasons to disable a tool platform-wide:

- Enforcing a security policy that prohibits a specific external integration
- Controlling costs from paid third-party APIs
- Compliance requirements restricting certain data egress paths

MCPs
----

The **MCPs** sub-tab manages Model Context Protocol servers — standardised tool endpoints that provide structured capabilities to Agentic Flow agents.

.. image:: /images/super-mcp.png
   :alt: MCP servers list
   :align: center
   :width: 70%

|

The MCP list displays:

- **MCP Name** — the name of the server
- **Status** — deployment state (Deployed)
- **Description** — what capabilities the MCP server provides
- **Enabled toggle** — whether agents can use this MCP
- **Timestamp** — when the MCP was last updated

Adding an MCP Server
~~~~~~~~~~~~~~~~~~~~

Click **+ Add MCP Server** to register a new MCP endpoint.

.. image:: /images/super-add-mcp.png
   :alt: Add MCP Server dialog
   :align: center
   :width: 70%

|

**Basic Information**

- **Name** — a descriptive label for the MCP server
- **Description** — what the server provides; shown to workspace admins and agent builders
- **Deployed Server URL** — the URL where the MCP server is reachable

**Configuration**

- **Docker Image URI with Tag** — the container image for the server (e.g., ``myregistry/mcp-server:latest``)
- **Port** — the port the server listens on
- **Environment Variables** — runtime configuration passed to the container
- **Arguments** — command-line arguments for the server process

Once added, the MCP server is available to Agentic Flow agents in all workspaces where it is enabled.

Best Practices
--------------

**Disable tools you don't use**
   Review the In-House Tools list and disable any tools that are not actively used by your workspaces. Enabled tools are available to every agent builder across the platform; disabling unused ones reduces the attack surface and avoids unintended external API calls.

**Security and compliance review for MCPs**
   Before adding a new MCP server, verify that the Docker image comes from a trusted registry and that the server URL is reachable only from the expected network path. Review MCP environment variables — they often contain API keys or credentials — and ensure they are rotated on the same schedule as other platform secrets.

**Document what each MCP provides**
   Write a clear Description when registering an MCP server. Agent builders rely on this description to understand what capabilities the MCP provides and whether it is appropriate for their use case.