> ## Documentation Index
> Fetch the complete documentation index at: https://docs.llmgrid.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP Servers

> Configure and manage MCP servers to enable tool-based interactions for models and agents.

## Overview

The **MCP Servers** section allows you to register, configure, and govern MCP servers that expose tools to models and agents. MCP servers enable models to call external tools in a controlled and auditable way.

MCP servers can be scoped by team and access group, monitored for health, and optionally published through the AI Hub for discovery.

***

## MCP Servers Page

At the top of the page, you can perform key management actions:

* **Add New MCP Server** – Register a new MCP server
* **Tabs**:
  * **All Servers** – View all configured MCP servers
  * **Connect** – Reference integration guidance and examples

***

## Filters & Context

### Current Team

Use the **Current Team** selector to filter MCP servers by team context. This determines which servers are visible and manageable.

### Access Group

Use the **Access Group** selector to filter MCP servers based on access group assignment.

***

## All Servers

The **All Servers** tab lists all MCP servers visible in the current context.

### Table Columns

Each server entry includes:

* **Server ID**\
  Unique identifier for the MCP server.

* **Name**\
  Human-readable server name.

* **Alias**\
  Optional alias used for routing or reference.

* **URL**\
  Base endpoint where the MCP server is hosted.

* **Transport**\
  Communication transport used by the server (for example, HTTP).

* **Auth Type**\
  Authentication mechanism required to access the server.

* **Health Status**\
  Indicates whether the server is reachable and healthy.

* **Access Groups**\
  Access groups allowed to interact with this server.

* **Created At / Updated At**\
  Timestamps for lifecycle tracking.

* **Actions**\
  Edit or manage the server configuration.

When no MCP servers are configured, an empty state message is displayed.

***

## Add New MCP Server

Select **Add New MCP Server** to register a server.

### Common Configuration Fields

* **Server Name** *(required)*\
  Friendly name used throughout the console and AI Hub.

* **Alias** *(optional)*\
  Short identifier that can be referenced programmatically.

* **Server URL** *(required)*\
  Base URL where the MCP server is hosted.

* **Transport**\
  Defines how requests are sent to the server.

* **Auth Type**\
  Specifies how requests are authenticated.

* **Access Groups**\
  Restrict which users, teams, or keys can access this server.

After saving, the server becomes available for selection in:

* Playground
* Agents
* Model tool configurations

***

## Health & Monitoring

Each MCP server tracks a **Health Status** that reflects connectivity and responsiveness.

Use health indicators to:

* Validate new servers after setup
* Investigate tool execution issues
* Monitor availability over time

***

## Connect Tab

The **Connect** tab provides implementation guidance for using MCP servers in API requests.

### Limiting Tools to MCP Servers

Requests can restrict which MCP servers or server groups are available by passing a dedicated request header.

This enables:

* Fine-grained tool control per request
* Safer production usage
* Reduced tool surface area

***

## Implementation Example

The **Implementation Example** section provides a complete request example demonstrating how to invoke tools exposed by MCP servers.

What the example shows conceptually:

* Sending a request to the responses endpoint
* Declaring MCP tools in the request payload
* Specifying server labels and URLs
* Providing required headers for server authentication
* Enforcing tool usage

This example is meant for reference and should be adapted to your application setup.

***

## MCP Servers in the AI Hub

MCP servers can be made visible in the **AI Hub**.

### Publishing MCP Servers

From the AI Hub:

1. Select **MCP Hub**
2. Choose **Select MCP Servers to Make Public**
3. Select one or more servers
4. Confirm publication

> Publishing a server only makes it discoverable. All authentication, access control, and guardrails remain enforced.

***

## Common Use Cases

* Connect models to internal systems or APIs
* Enable function-like tool execution
* Support agent-based workflows
* Standardize external integrations behind a governed interface
* Provide discoverability through the AI Hub

***

## Best Practices

* Use access groups to restrict server visibility
* Monitor health status regularly
* Validate servers in Playground before production use
* Avoid overexposing tools; publish only approved servers
* Document server capabilities via AI Hub links

***

## Related Sections

* **Agents** – Use MCP servers as tools
* **Models** – Attach MCP tools to model workflows
* **Playground** – Test MCP tool execution
* **AI Hub** – Publish MCP servers for discovery
* **Virtual Keys** – Enforce access and limits