@three-ws/billing-mcp
An AI agent's own account economics over MCP, plan quotas, metered usage, invoices, receipts, and earnings. Read-only and account-scoped.
A Model Context Protocol server that lets an AI agent self-serve its own account economics over stdio: how much of its plan quota is left, what its metered usage rolled up to this period, the line-item invoice and per-charge receipts behind every charge, and the earnings its agents have made, all without pulling in the human account owner.
Every read hits the live three.ws billing API: nothing is mocked. The account-scoped reads run against the account you own, resolved from your three.ws session. This server is read-only: it surfaces private billing data but never signs, charges, or moves funds.
Install
npm install @three-ws/billing-mcp
Or run with npx (no install):
npx @three-ws/billing-mcp
Quick start
Claude Code, one line:
claude mcp add billing -- npx -y @three-ws/billing-mcp
Claude Desktop / Cursor (claude_desktop_config.json or mcp.json):
{
"mcpServers": {
"billing": {
"command": "npx",
"args": ["-y", "@three-ws/billing-mcp"],
"env": {
"THREE_WS_SESSION": "<your __Host-sid cookie>"
}
}
}
}
THREE_WS_SESSION is required for every read except the public fee rate, it's the value of the __Host-sid cookie from a signed-in three.ws browser session, and the API uses it to resolve your account and return only your data.
Inspect the surface with the MCP Inspector:
npx -y @modelcontextprotocol/inspector npx @three-ws/billing-mcp
| Tool | Type | What it does |
|---|
get_billing_summary | read · session | Your plan tier, its quota ceilings, and live usage against them (avatars + storage, agents, MCP calls, LLM calls) + headroom. |
query_usage | read · session | Metered usage rolled into an invoice statement for a period, per-action line items, totals, and reconciliation. |
export_billing_history | read · session | The same line items as a ready-to-save CSV payload (with a suggested filename and a parsed preview). |
get_receipt | read · session | One itemized receipt, per-charge by event_id (with settlement tx + explorer link) or a signed skill-purchase receipt by purchase_id. |
get_fee_info | read · public | The current platform fee rate (fee_bps + fee_percent). Needs no session. |
Everything reads live state: usage and invoices move between calls, so no tool is idempotent, and nothing here mutates anything.
get_billing_summary: none.
query_usage: period (YYYY-MM, optional), from / to (ISO-8601, optional; ignored when period is set). Defaults to the current UTC calendar month.
export_billing_history: period, from, to (same as query_usage), preview_rows (0-100, default 5, how many parsed rows to echo in preview; the full CSV is always returned).
get_receipt: exactly one of event_id (numeric, per-charge receipt) or purchase_id (UUID, signed skill-purchase receipt).
get_fee_info: none.
Examples
> {}
{
"ok": true,
"plan": "free",
"quotas": {
"max_avatars": 10,
"max_bytes_per_avatar": 26214400,
"max_total_bytes": 262144000,
"mcp_calls_per_day": 1000
},
"usage": {
"avatar_count": 1,
"total_bytes": 1234540,
"agent_count": 1,
"mcp_calls_24h": 0,
"llm_calls_month": 0
},
"remaining": { "avatars": 9, "total_bytes": 260909460, "mcp_calls_today": 1000 }
}
> {}
{
"ok": true,
"period_label": "2026-08",
"period": { "from": "2026-08-01T00:00:00.000Z", "to": "2026-09-01T00:00:00.000Z" },
"line_item_count": 0,
"line_items": [],
"totals": {
"charge_count": 0,
"gross_atomics": "0", "fee_atomics": "0", "net_atomics": "0",
"gross_usd": "0.00", "fee_usd": "0.00", "net_usd": "0.00",
"currency": "USDC"
},
"reconciliation": { "total": 0, "reconciled": 0, "unreconciled": 0, "all_reconciled": true }
}
> { "preview_rows": 5 }
{
"ok": true,
"filename": "three-ws-invoice-2026-08.csv",
"content_type": "text/csv; charset=utf-8",
"row_count": 1,
"preview": [ { "action": "TOTAL", "count": "0", "gross_usd": "0.00", "fee_usd": "0.00" } ],
"csv": "action,label,count,units,gross_usd,fee_usd,discount_bps\nTOTAL,,0,,0.00,0.00,\n"
}
> {}
{ "ok": true, "fee_bps": 250, "fee_percent": "2.5" }
A quiet account reads like the statements above: zeroed totals with
all_reconciled: true. That is the healthy empty state, not a failed call. Every
read without THREE_WS_SESSION (except get_fee_info) fails loudly instead of
returning empty data:
{ "ok": false, "error": "no_session", "status": 401,
"message": "/api/billing/summary is account-scoped and needs your three.ws session. …" }
What you owe
- "What did this cost me?" →
query_usage / export_billing_history / get_receipt, metered charges, the period statement, and the receipt behind any single charge.
get_billing_summary sits above them: your plan, its quota ceilings, and how much headroom is left.
Money & units
- Charge amounts are in USDC atomics (6 decimals) alongside a human
*_usd string, e.g. gross_atomics: "150000" is gross_usd: "0.15".
reconciliation on the usage reads tells you whether every metered charge maps to a real on-chain settlement (all_reconciled, plus counts).
Requirements
- Node.js >= 20.
- Network access to
https://three.ws (or your own THREE_WS_BASE).
Environment variables
| Variable | Required | Default |
|---|
THREE_WS_BASE | no | https://three.ws |
THREE_WS_TIMEOUT_MS | no | 20000 |
THREE_WS_SESSION | yes (all reads but get_fee_info) | - |
THREE_WS_SESSION is the value of the __Host-sid cookie from a signed-in three.ws browser session. Treat it like a password, it grants read access to your private billing data.
Links
Part of the three.ws SDK suite, 3D AI agents, on-chain identity, and agent payments.
Website · Changelog · GitHub