MCP: Connect via integration token

Overview

The Quantum Metric MCP Server connects your Quantum Metric subscription to AI assistants and automated systems through the Model Context Protocol (MCP), an open standard for giving AI tools structured access to external systems.

Once connected, an AI client can query behavioral data, investigate sessions and errors, run funnels and saved queries, and trigger the workflows and monitoring agents you have built in Quantum Metric.

This guide covers integration tokens: credentials that let an automated system call Quantum Metric MCP with no person present. Use them for scheduled analysis, experiment read-outs, automated remediation and agent frameworks.

What an integration token is

An integration token is a credential for you subscription. Your automation sends it with every request.

Before you begin

You need:

• An active Quantum Metric subscription.
• Your subscription name, as it appears in your Quantum Metric URL.
• An MCP-capable client: Claude Code, Claude (desktop and web), ChatGPT, Cursor, VS Code, Gemini CLI, or any client that implements MCP.

Endpoint

There is one production endpoint, with your subscription in the path:

https://api.quantummetric.com/mcp/<subscription>

Replace <subscription> with your subscription name.

Getting an integration token

MCP Tokens are created by Quantum Metric. Talk to your Customer Success representative to get one for your subscription.

Connect with an integration token

Send the token as a bearer credential to the endpoint with your subscription in the path.

Authorization: Bearer <your-integration-token>

Claude Code:

claude mcp add --transport http --header "Authorization: Bearer <your-integration-token>" felix https://api.quantummetric.com/mcp/<subscription>

Clients that take JSON configuration:

{
  "mcpServers": {
    "felix": {
      "type": "http",
      "url": "https://api.quantummetric.com/mcp/<subscription>",
      "headers": {
        "Authorization": "Bearer <your-integration-token>"
      }
    }
  }
}

Note: If you get a 400 error, ensure you are sending with the correct subscription.

Verify the connection

1. In Claude Code, run claude mcp list. The felix server should show as connected. Other clients show connection status in their MCP or connector settings.
2. Ask a question that needs your data, such as "List the dashboards in this subscription."
3. Call get_context and confirm the subscription is the one you expect.

Set your workspace and timezone on every call

A token session has no user profile, so it has no default workspace and runs in UTC. Pass both values on every call.

• Workspace is required. Team workspaces, and any segment attached to them, scope what a query sees. A data call without workspace fails with Error: no workspace set. Pass the workspace parameter (list_workspaces shows the options).
• Timezone defaults to UTC. Pass timezone as an IANA name, such as America/New_York, so "yesterday" and daily breakdowns line up with your business day.

If automated numbers disagree with numbers a colleague pulled in the Quantum Metric UI, compare the workspace and timezone first. The UI uses their workspace and their own timezone.

Step 1: find the workspace ID. Call list_workspaces; it returns each workspace's id and name. Or resolve a name directly:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "switch_workspace",
    "arguments": { "workspace": "Checkout Team" }
  }
}

The reply gives the exact workspace value to pass. It changes nothing on the server.

Step 2: pass workspace and timezone on each call. For example, a run_query call:

{
  "jsonrpc": "2.0",
  "id": 2,
  "method": "tools/call",
  "params": {
    "name": "run_query",
    "arguments": {
      "query": "Conversion rate by device type",
      "time_phrase": "yesterday",
      "workspace": "<workspace-id>",
      "timezone": "America/New_York"
    }
  }
}

The same two arguments work on run_saved_query, run_funnel, execute_dashboard_card and the other data tools. create_saved_query is the exception: it takes workspace_id.

If an AI agent makes the calls, put the values in its instructions, for example:

On every Felix tool call that accepts them, pass workspace: "<workspace-id>" and timezone: "America/New_York".

Step 3: check what answered. Workspace-scoped responses echo the workspace that answered. run_saved_query also returns workspace_id, workspace_name and workspace_filter_applied.

Building autonomous workflows

An integration token makes calls as your subscription's service account. The automation's access does not depend on any person's session or employment.

Know what each call does

Before you put an invoke tool in an unattended loop, confirm what that agent or workflow does when it runs.

Confirm your scope

• run_saved_query returns a context block with workspace_id, workspace_name and workspace_filter_applied. workspace_filter_applied: false means the workspace has no segment filter; the workspace still applies.
• To confirm empirically, count a value your workspace segment excludes in All Traffic and in the workspace, and compare.

Handle failures by status code

• 400: configuration error in the URL. Retrying will not fix it.
• 401: credential problem. Check the token value and replace it if revoked.
• 403: the token is valid but missing Felix MCP access. Alert a person; retrying will not fix it.

Troubleshooting

Getting help

Contact your Customer Success Engineer for help connecting MCP to your systems or enabling token access settings.

Appendix: tool reference

The server typically exposes tools across 16 capability areas. The list may change as the product develops; your client retrieves the current list at connect time. Some per-subscription tools, such as Voice of Customer survey tools, are unavailable to integration tokens.