Add Real Business Trust Signals to Claude Desktop in 60 Seconds

#mcp #ai #claude #tutorial

Reviews are fake. Ratings are gamed. What if your AI could check verified behavioral data instead?

This is a 60-second setup. No code. No install. Just add two lines to your MCP config.

What you get

Ask Claude, Cursor, or Windsurf:

"How trustworthy is Equinor?"

And get:

EQUINOR ASA (923609016)
Operating for 53.5 years (founded 1972-09-18)
21,408 employees
Revenue: 72,543M NOK
Equity ratio: 37.6%

Commitment signals:
  Temporal (longevity): 95/100
  Financial (health): 90/100
  Operational (activity): 95/100
  Overall commitment: 93/100

This is not a review. This is structural data — years operating, employee count, financial ratios — sourced from Brønnøysundregistrene, the Norwegian government business register. It cannot be gamed. It cannot be bought.

The setup (60 seconds)

Claude Desktop — edit ~/.claude/claude_desktop_config.json:

{
  "mcpServers": {
    "proof-of-commitment": {
      "type": "streamable-http",
      "url": "https://poc-backend.amdal-dev.workers.dev/mcp"
    }
  }
}

Cursor — go to Settings → MCP, add:

{
  "proof-of-commitment": {
    "type": "streamable-http",
    "url": "https://poc-backend.amdal-dev.workers.dev/mcp"
  }
}

That is it. Restart the app. Done.

What you can ask

Once connected, try:

"How trustworthy is Equinor?" → full commitment profile
"Look up Norwegian business by org number 923609016" → direct lookup
"Is Avinor a reliable company?" → Norwegian state-owned airports operator

The MCP server exposes three tools:

Tool	What it does
`query_commitment`	Query behavioral data for any domain
`lookup_business`	Search by business name
`lookup_business_by_org`	Lookup by 9-digit org number

Why this matters

The entire trust data market — $10B+ annually (Verisk, D&B, FICO, Trustpilot) — is built on opinions and declarations. Nobody sells AI what actually happened: how long a business has operated, how many employees stayed, whether their financials are solid.

This is a narrow wedge: Norwegian businesses, government data. But the architecture is the point. Behavioral commitment data is structurally harder to fake than any review. That is the primitive that is missing from how AI evaluates trust.