Foglight MCP Server

The Foglight MCP Server enables integration between Foglight and AI-powered development tools through the Model Context Protocol (MCP). This allows developers and administrators to interact with Foglight monitoring data directly from MCP-compatible AI clients, such as VS Code with GitHub Copilot, Cursor, Claude Desktop, or ChatGPT.

This section covers the following key areas:

• Overview
  • What is MCP?
  • Why Use MCP with Foglight?
  • What Does the Foglight MCP Server Provide?
  • Tool Categories
    • Alarms
    • Agents and Agent Managers (FglAM)
    • Performance Investigation (PI)
    • TopSQL (Cross-Database)
    • Topology / GraphQL
• Before You Start
  • Authentication
  • Prerequisites
• Configuring Your AI Client
  • VS Code (GitHub Copilot)
  • Cursor
  • Claude Desktop
  • ChatGPT
• Signing In
• Validating the Connection
• Troubleshooting
  • 401 Unauthorized Error
  • 403 Forbidden Error
  • Tools Return Empty Results
  • OAuth Redirect URI Rejected (HTTP 400)
• Current Limitations
• Frequently Asked Questions

Overview

This section explains what MCP (Model Context Protocol) is and what the Foglight MCP Server provides. It is written for readers who are already familiar with Foglight concepts (FMS, agents, alarms, PI/Performance Investigation, Top SQL), but are new to MCP.

What is MCP?

MCP (Model Context Protocol) is an open protocol that lets an AI client (for example, an assistant built into an IDE or chat tool) call tools exposed by a server.

In practice, MCP provides:

• A standard way for an AI client to discover what tools are available
• A standard way to call those tools with parameters
• A standard way to return results back to the AI client (usually as structured JSON)

Think of MCP as a “connector” layer between an AI assistant and operational systems like Foglight.

Why Use MCP with Foglight?

Foglight already contains rich monitoring and performance data. MCP makes that data easier to use in interactive workflows where you want to:

• Ask questions in natural language (for example: “What are the top SQL statements by impact?”)
• Retrieve supporting evidence quickly (alarms, Top SQL, PI breakdown)
• Drill down iteratively without manually switching between many screens

MCP does not replace the Foglight UI. It provides a programmatic interface that AI clients can use to help you navigate Foglight data faster.

What Does the Foglight MCP Server Provide?

The Foglight MCP Server is an MCP server that exposes Foglight data through a set of tools.

Typical capability areas include:

Exact tool names and parameters depend on the server build and deployment configuration, but the usage pattern is consistent.

Tool Categories

Alarms

Use these when you want to answer “What is wrong right now?” or “What happened last week?”

Examples of questions you can ask:

• “Show me current critical alarms.”
• “Do we have any active alarms for this target?”
• “What are the longest-running alarms right now?”
• “Show alarms created in the last hour.”
• “Give me an alarm summary grouped by agent.”

Agents and Agent Managers (FglAM)

Use these when you want to understand what Foglight is monitoring and whether collection infrastructure looks healthy.

Examples of questions you can ask:

• “List monitored agents (active vs inactive).”
• “Show agents for domain MSSQL (or POSTGRES, ORACLE).”
• “Which agent managers are disconnected?”
• “How many agents are connected to each agent manager?”
• “Is this agent active?”

Performance Investigation (PI)

Use these when you want to diagnose database performance.

Examples of questions you can ask:

• “Show PI highlights for this database instance.”
• “Where is this database instance spending time (CPU vs Lock vs I/O)?”
• “Show top SQL by I/O on this database instance.”
• “Show top SQL by Lock on this database instance.”
• “Show blocking history for this database instance.”
• “Show top users or top databases by workload on this database instance.”

TopSQL (Cross-Database)

Use this when you want the most impactful statements across all monitored databases.

Examples of questions you can ask:

• “Show top SQL across all databases by impact.”
• “Show top SQL for domain MSSQL.”
• “Show top SQL for this target.”
• “Get full SQL text for the top statement (if available).”

Topology / GraphQL

Use GraphQL when you need access to Foglight topology properties/metrics not covered by the higher-level tools.

Examples of questions you can ask:

• “Show CPU utilization for this host.”
• “Show DBSS wait statistics for this database instance.”
• “Find the topology property/metric name for ‘cpu’ in the SQL Server domain.”

Before You Start

Authentication

• Foglight MCP Server tools are intended for read-only retrieval of monitoring and performance data.
• Authentication is based on OAuth2. You need to authenticate to Foglight Cloud to get access.
• The server should be accessed according to your organization’s security policies.

Prerequisites

Before you start, ensure you have the following:

Configuring Your AI Client

The Foglight MCP Server supports any MCP-compatible AI client that can connect over HTTP. Below are configuration examples for the most common clients.

VS Code (GitHub Copilot)

To add the MCP server to VS Code:

1. In your workspace, create or open the file .vscode/mcp.json.
2. Add a server entry that points to your Foglight MCP endpoint:

{
  "servers": {
    "foglight-mcp-sso": {
      "type": "http",
      "url": "https://<FOGLIGHT_MCP_ENDPOINT>/mcp"
    }
  }
}

Cursor

Create or edit .cursor/mcp.json (project scope) or ~/.cursor/mcp.json (global scope):

{
  "mcpServers": {
    "FMS MCP Server": {
      "type": "http",
      "url": "https://<FOGLIGHT_MCP_ENDPOINT>/mcp",
      "auth": {
        "CLIENT_ID": "foglight-mcp",
        "scopes": ["openid", "profile", "email"]
      }
    }
  }
}

Authentication is handled via OAuth2. When you first connect, Cursor opens the identity provider login page automatically.

Claude Desktop

Open claude_desktop_config.json via Settings > Developer > Edit Config and add:

{
  "mcpServers": {
    "FMS MCP Server": {
      "url": "https://<FOGLIGHT_MCP_ENDPOINT>/mcp"
    }
  }
}

Save the file and fully restart Claude Desktop. The Foglight tools will appear under the hammer icon in the chat input.

ChatGPT

ChatGPT supports MCP servers over HTTP only. The Foglight MCP endpoint must be reachable from the internet.

To connect:

1. Enable developer mode in Settings → Apps → Advanced settings.
2. Click Create app and enter:
   • Name: foglight-mcp
   • MCP Server URL: https://<FOGLIGHT_MCP_ENDPOINT>/mcp
   • Authentication: OAuth
3. Under Advanced settings, select User Defined OAuth Client and configure the OAuth endpoints for your identity provider (Auth URL, Token URL, client ID foglight-mcp, default scopes openid, profile, email).
4. Click Create.
5. In a new chat, select Developer Mode from the + (tools) menu and choose the Foglight app.

For the full OAuth configuration fields, publishing workflow, and admin controls, refer to the OpenAI MCP documentation and ChatGPT developer mode guide.

Signing In

When you use a Foglight MCP tool for the first time, VS Code opens an interactive sign-in flow automatically.

To sign in:

1. In VS Code, try a Foglight MCP tool call (for example, “show current alarms”).
2. If you are prompted to sign in, complete the browser sign-in.
3. If prompted to provide a client ID, enter foglight-mcp.
4. If asked to provide an OAuth client secret, leave blank.
5. Run the tool call again after sign-in completes.

Validating the Connection

To validate that the MCP server is working correctly, try one of these simple read-only checks in your AI client:

• “Show current alarms”
• “List monitored agents”
• “Show top SQL across all databases by impact”

If SSO is working, you should receive results after signing in.

Troubleshooting

This section provides solutions for common issues when connecting to the Foglight MCP Server.

401 Unauthorized Error

Cause

This error typically occurs when:

• You are not signed in via SSO for this endpoint
• Your account is not authorized for Foglight MCP
• The endpoint URL is incorrect (missing /mcp or wrong host)

Resolution

1. Open https://<FOGLIGHT_MCP_ENDPOINT>/mcp in a browser and confirm you can sign in.
2. If the browser sign-in works but VS Code still fails, contact your support channel for Foglight MCP access.

403 Forbidden Error

Cause

Your user or group is not allowed to access Foglight MCP.

Resolution

Request access to the Foglight MCP application/role in your IdP.

Tools Return Empty Results

Cause

This issue typically occurs when:

• There is no data for the requested timeframe
• The requested domain/instance is not being monitored or PI is not enabled

Resolution

1. Try a broader query (for example, “top SQL across all domains”).
2. Confirm the target is monitored in Foglight.

OAuth Redirect URI Rejected (HTTP 400)

Cause

During OAuth sign-in, your AI client sends a redirect URI back to the identity provider (for example, Keycloak). Each client uses its own native callback URI. If that URI is not registered as an allowed redirect URI on the foglight-mcp OAuth client, the identity provider rejects the request with:

HTTP 400 — Invalid parameter: redirect_uri

Resolution

Contact your Quest account manager and request that your AI client’s redirect URI be added to the allowed redirect URIs for the foglight-mcp OAuth client. Include the following in your request:

• The exact redirect URI your client sends (check your client’s documentation or error logs)
• The Foglight MCP Server endpoint you are connecting to

No wildcard is needed — provide the exact URI.

Current Limitations

The following capabilities are not currently supported by the MCP Server:

• Ability to retrieve details from ad-hoc queries run by Foglight (such as show current sessions)
• PI (Performance Investigator) information for:
  • Oracle databases
  • MySQL databases
  • SQL Server Analysis Services (SSAS) domains

These limitations may be addressed in future releases.

Frequently Asked Questions

Does MCP mean Foglight data is sent to a public AI service?

Not necessarily. MCP describes how an AI client calls tools. Where the AI model runs and what data is shared depends on your deployment and the client configuration.

Should I use GraphQL for everything?

Usually no. Prefer the higher-level tools (Alarms, PI, TopSQL) for common operations. Use GraphQL for advanced/custom data retrieval.

What if a tool returns an ID or compressed SQL text?

That is expected in some workflows. Use the returned IDs to call the next tool in the chain. Use the decompress tool when a response provides compressed/encoded SQL text.