@quetra/mcp

MCP server for QuetraAI — expose AI agent spending governance as Model Context Protocol tools. Works with Claude Desktop, Claude Code, and any MCP-compatible client.

Two Ways to Connect

| Method | Transport | Use Case | | ------------------------ | --------------- | ------------------------------------------------------------------------ | | Local (this package) | stdio | Claude Desktop config, npx, programmatic embedding | | Remote | Streamable HTTP | One URL, multi-agent, no install: https://mcp.quetra.dev/<api-key>/mcp |

The remote server supports multiple agents per connection — see Remote MCP Setup below.

Install

npm install @quetra/mcp

Claude Desktop Setup (Local)

Add to your claude_desktop_config.json:

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

Remote MCP Server

No install required. Add the remote server URL directly in Claude Code or Claude Desktop:

https://mcp.quetra.dev/<your-api-key>/mcp

The remote server is multi-agent — use quetra_list_agents to discover available agents, then pass agentId to each tool call. One connection governs all your agents.

Standalone Usage

Run directly from the command line (stdio transport):

QUETRA_API_KEY=sk_... QUETRA_AGENT_ID=agent-uuid npx @quetra/mcp

Programmatic Usage

Embed the MCP server in your own application:

import { createQuetraMCPServer } from "@quetra/mcp";

const server = createQuetraMCPServer({
  apiKey: "sk_...",
  agentId: "agent-uuid",
  gatewayUrl: "https://gateway.quetra.dev",
});

Available Tools

| Tool | Description | | ---------------------- | ----------------------------------------------------------------------------------------------------------------- | | quetra_evaluate | Pre-flight spend check — evaluate a payment against mandate rules and budget. Returns per-rule pass/fail details. | | quetra_check_budget | Budget status: total, spent, remaining, per-transaction limit, reset schedule. | | quetra_can_spend | Spend check with rejection reasons — shows exactly which rules failed. | | quetra_fetch | x402-aware fetch — make HTTP requests that auto-handle payment-required responses. | | quetra_transactions | List recent transaction history with optional filters. | | quetra_acp_checkout | Initiate a Stripe ACP merchant checkout flow. |

The remote server adds quetra_list_agents — discover available agents in your organization.

Environment Variables

| Variable | Required | Description | | -------------------- | -------- | ------------------------------------------------------------------- | | QUETRA_API_KEY | Yes | Your organization's API key (sk_...) | | QUETRA_AGENT_ID | Yes | The agent's UUID (local only — remote uses agentId per tool call) | | QUETRA_GATEWAY_URL | No | Gateway URL (defaults to https://gateway.quetra.dev) |

Getting Started

1. Sign up at app.quetra.dev
2. Create an organization and register an agent
3. Create a mandate with spending rules and budget
4. Generate an API key
5. Add the MCP server config to Claude Desktop (local) or paste the remote URL

Related Packages

• @quetra/sdk — Client SDK for programmatic agent integration
• @quetra/core — Types and rule engine

License

MIT