Labs Connect MCP server

The Signal Line

Puts your customer data behind an MCP server, so an AI assistant answers questions inside the tools your team already has open.

Connect Claude or any MCP client and ask about the feedback in plain English. It can search comments, pull theme trends and segment health, and flag an account for review.

WHAT IS MCP

Tools an AI assistant can actually use

The Model Context Protocol (MCP) is an open standard that lets an AI assistant like Claude call an application’s own tools and read its data directly, instead of a human copy-pasting context back and forth. The Signal Line is a real MCP server, and every tool, resource and prompt below is read from it live each time this page loads.

LIVE CATALOGUE

Tools, resources & prompts

Tools (5)

search_feedback

Hybrid (semantic + lexical) search over customer feedback. Optional filter.

query string required What to search for
segment string one of: Enterprise, Mid-Market, SMB
account_ref string Limit results to one account

get_theme_trends

Counts of feedback by theme and sentiment. Optional segment filter.

segment string one of: Enterprise, Mid-Market, SMB

get_segment_health

Average account health and usage trend for a segment.

segment string required one of: Enterprise, Mid-Market, SMB

flag_account_for_review

Flags an account for review. Fails on an unknown account_ref.

account_ref string required
reason string required

get_flagged_accounts

Lists accounts flagged for review, most recent entry per account.

No parameters.

Resources (2)

VoC segment definitions

Definitions for customer segments used across the signal_line tools.

voc://segments

VoC persona definitions

Definitions for personas used across the signal_line tools.

voc://personas

Prompts (3)

weekly_cs_brief

Prioritised weekly brief: which accounts need attention and why.

segment Limit the brief to one segment (Enterprise, Mid-Market, SMB)

account_deep_dive

Narrative summary of one account: recent feedback, flag status, recommended next step.

account_ref required The account_ref to investigate

churn_risk_scan

Identifies which segment(s) are trending negative and what to investigate first.

segment Limit the scan to one segment (Enterprise, Mid-Market, SMB)

CONNECT YOUR OWN CLIENT

Wire this into Claude

URL https://www.cxmatters.com.au/mcp

Transport Streamable HTTP

Header Authorization: Bearer <your-token>

First, a demo token

This is a live demo rather than a public endpoint, so the config below needs a token before it will connect. Ask for one and you will get a read-only key that can query every tool above.

Ask for a demo token →

Then, Claude Desktop

Claude Desktop’s built-in Custom Connector UI only supports OAuth Client ID and Secret, not a static bearer token, so this server connects through the mcp-remote stdio bridge instead. It runs over npx, so Node needs to be installed.

{
  "mcpServers": {
    "signal_line": {
      "command": "npx",
      "args": [
        "mcp-remote@latest",
        "https://www.cxmatters.com.au/mcp",
        "--header",
        "Authorization:${AUTH_HEADER}"
      ],
      "env": {
        "AUTH_HEADER": "Bearer <demo_token>"
      }
    }
  }
}

Paste that into Settings → Developer → Edit Config, save, and restart Claude Desktop. The tools appear under the plug icon in a new chat.

Access today is gated by a shared bearer-token header, checked against an environment secret: a demo-grade shortcut, not a production auth model. The demo token is read-only, so it can query tools freely but any write (like flagging an account) is politely refused. A real deployment would sit behind Doorkeeper or OAuth 2.1, issuing scoped, revocable tokens per client instead of one shared secret per role.

RECENT ACTIVITY

Recent queries through the Signal Line

Quiet lately. The most recent call through this server was 17 days ago. The endpoint is live either way: the tools above are read from it on every page load.

The last 8 tool calls logged by the server.
Tool	Input	Status	Duration	Token	When
get_segment_health	`{"segment": "Enterprise"}`	success	75ms	demo	17 days ago
get_segment_health	`{"segment": "Enterprise"}`	success	9ms	demo	about 2 months ago
flag_account_for_review	`{"reason": "health check", "account_ref": "acc_1"}`	error	0ms	demo	about 2 months ago
search_feedback	`{"query": "webhook reliability problems", "segment": "Enterprise"}`	success	299ms	demo	about 2 months ago
search_feedback	`{"query": "Vantage Meridian", "segment": "Mid-Market"}`	success	158ms	demo	3 months ago
search_feedback	`{"query": "acct_1 support ticket"}`	success	767ms	demo	3 months ago
search_feedback	`{"query": "Vantage Meridian Solutions"}`	success	154ms	demo	3 months ago
get_flagged_accounts	`{}`	success	7ms	demo	3 months ago

An error on flag_account_for_review is the demo token being refused a write. That is the behaviour described below, not a fault.

Other labs

The same customer feedback runs through each of these.

See