Loading
Loading
Burnbound is a spending control plane for AI agents. Before an agent refunds a Stripe charge, pays with a Stripe Issuing card or pays an x402 or MPP API, Burnbound checks it against rules a person set: per-payment and daily caps, team budgets, custom rules and human approval above a threshold. Burnbound never holds funds, wallet keys or the customer's Stripe secret key: refunds run in an action gateway the customer hosts, card money stays in the customer's Stripe Issuing balance, and API payments are signed by a wallet the customer brings.
From creating an agent key in the dashboard to the agent's first fetch_paid call, in Claude Code or Cursor.
Policies, per-payment and daily caps, allowed hosts, custom spending rules, subteams and organization rules, human approvals, signing modes and the x402 v2 payment flow.
Configuration, the fetch_paid, get_budget, propose_rule, list_payments, get_approval_status, search_paid_apis, diagnose_payment, stripe_refund, stripe_credit, shopify_refund and task mandate tools, approval from the chat, error codes, and the wallet and doctor CLIs of @burnbound/mcp.
Status of the JavaScript SDK (payFetch and wrapFetchWithPayment): not published on npm yet; use the MCP server today.
What Burnbound stores and sees in each signing mode, protection against injected payments, the cooperative limit of client-side signing, and out-of-policy detection.
Your own business logic in every payment decision: what Burnbound sends to your endpoint, how to answer and verify the signature, failures, approvals, cards and secret rotation.
How Burnbound decides a support agent's refunds and credits: the facts the action gateway reads from Stripe or Shopify, the rule conditions on them, the refund templates and packs, and the three actions.
Approve a whole task once instead of each payment: what a task mandate covers, what it never lifts, and how the agent asks for one with request_mandate.
Managed and observed spend, token cost from Anthropic and OpenAI, cost centers and the monthly accounting export, side by side for each agent.
A ZIP of decisions, approvals, policy changes and access for security, compliance or an auditor, with a SHA-256 manifest you can verify against the audit log.
The methodology of the savings report: what was avoided for sure, the spend your limits contained, and what never counts.
burnbound-mcp audit-refunds: replay your Stripe refunds of the last days against a Burnbound policy on your own machine, with a read-only key, and see what it would have denied or held.
Short answers about wallets, networks, fees, plans, custody and supported agents.
Step-by-step walkthroughs are in the guides, and the exact setup for Claude Code, Cursor, Claude Desktop, the Vercel AI SDK, LangChain.js and Mastra is in the integrations.
For LLMs and agents: /llms.txt lists every page, and /llms-full.txt has all of them as one Markdown file.