QuetraAI + Claude Cowork Setup

Connect QuetraAI spending governance to Claude Desktop or Claude Code via MCP. Two connection options — remote (recommended) or local.

Prerequisites

• Claude Desktop or Claude Code installed
• A QuetraAI account at app.quetra.dev
• Node.js 18+ (only if using local setup)

Step 1: Create Your Organization

1. Go to app.quetra.dev/auth/signup
2. Create an account (email/password or social login via GitHub, Google, Apple, or Microsoft)
3. Create an organization — this is your governance namespace

Step 2: Register an Agent

QuetraAI uses a 1:1 agent-to-mandate model — create one agent per task, then assign it a single mandate.

1. Navigate to Agents in the sidebar
2. Click Register Agent
3. Fill in name, description (critical!), framework, and execution mode
4. Copy the Agent ID (UUID)

Step 3: Create a Mandate

Without an active mandate, all spending requests are rejected. Configure budget, rules, and expiration.

1. Navigate to Mandates → Create Mandate
2. Select your agent, set spending limit, per-transaction limit, and budget reset period
3. Add rules: category allowlist, rate limit, vendor restrictions
4. Click Create Draft, then Activate

Step 4: Generate an API Key

1. Navigate to API Keys → Create API Key
2. Copy the key immediately (starts with sk_, only shown once)

Step 5: Connect to Claude

Option A: Remote Server (recommended)

No install required. Add the remote MCP server URL:

Claude Code: Settings → MCP Servers → Add Remote Server

https://mcp.quetra.dev/sk_your_api_key_here/mcp

Claude Desktop: Settings → Connectors → Add → paste the URL above.

The remote server is multi-agent — use quetra_list_agents to discover agents. Claude reads descriptions to pick the right one.

Option B: Local Server (stdio)

Add to your Claude Desktop config:

{
  "mcpServers": {
    "quetra": {
      "command": "npx",
      "args": ["@quetra/mcp"],
      "env": {
        "QUETRA_API_KEY": "sk_your_api_key_here",
        "QUETRA_AGENT_ID": "your-agent-uuid-here",
        "QUETRA_GATEWAY_URL": "https://gateway.quetra.dev"
      }
    }
  }
}

Step 6: Verify the Integration

Restart Claude and try these prompts:

• "Check my QuetraAI budget status" — calls quetra_check_budget
• "Can I spend $2.00 on research at firecrawl.dev?" — calls quetra_can_spend
• "Can I spend $500 on entertainment at netflix.com?" — should be rejected
• "Show my recent QuetraAI transactions" — calls quetra_transactions

How It Works

Claude Cowork
    │
    ├─ User: "Buy the premium dataset from data.world for $15"
    │
    ├─ Cowork calls: quetra_evaluate(amount: 1500, vendor: "data.world", category: "research")
    │       │
    │       ▼
    │   MCP Server → @quetra/sdk → POST gateway.quetra.dev/api/v1/gateway/evaluate
    │       │
    │       ▼
    │   QuetraAI Gateway:
    │     1. Authenticate API key
    │     2. Resolve active mandate for agent
    │     3. Evaluate ALL rules (AND logic)
    │     4. Atomically decrement budget
    │     5. Write to audit log
    │     6. Fire webhooks (async)
    │     7. Return decision
    │
    └─ Cowork: "Approved! You have $8.50 remaining in your weekly budget."

Troubleshooting

Next Steps

• Monitor activity on the dashboard
• Set up webhooks for budget warnings at 80%
• MCP Overview & Tools