Spend Control for Claude Code
Claude Code is the most popular autonomous coding agent — and it buys things: compute, API credits, even payment rails via x402. A prompt saying 'be careful' isn't a spending policy.
Why Claude Code agents overspend
Long-running coding sessions call tools thousands of times; any paid tool in a retry loop multiplies spend.
Claude Code can trigger payments through agentic rails — one wrong tool call settles real money.
Provider limits are per-provider; nothing governs the cross-provider, cross-merchant whole.
How it works
Add sipi.bot as an MCP tool to your Claude Code config. Before any spend, Claude calls sipi_guard with the amount, merchant, and category and receives APPROVED, BLOCKED, or FLAGGED.
Rules that fit coding workflows
Per-transaction cap on compute and API purchases.
Merchant allowlist for known vendors — unknown GPU/compute sellers get blocked.
Velocity limit so a retry loop dies in seconds.
Config snippet (MCP)
{
"mcpServers": {
"sipi": {
"command": "npx",
"args": ["-y", "@sipi/bot-mcp"]
}
}
}Claude Code discovers the tool on startup. See /tutorials/integrate-with-claude-code for the full walkthrough.
FAQ
Does sipi.bot slow Claude Code down?
No — decisions take about 5 ms and the tool is called only on spend actions, not on every turn.
Can it block a specific merchant?
Yes — the merchant allowlist blocks any vendor you haven't approved, like an unknown GPU seller.
Is this the same as Claude's provider limits?
No. Provider limits cap spend at one provider and usually report after the fact. sipi.bot governs every transaction across every merchant before money moves.
Related
Stop the next $12,400 night.
One API call (or MCP tool) in front of every agent transaction — APPROVED, BLOCKED, or FLAGGED, deterministic, ~5 ms, fully logged.
See plans — from $99/mo Try a live check