Skip to content

Security

The server has no private keys and makes no blockchain transactions. Its tools create invoices and read data through the Request Network API. A receive-only address can’t be drained, so the worst a misbehaving assistant can do is create unwanted invoices.

Before any network call, every input is validated:

  • Amounts: plain positive decimals only. 0, negatives, exponents and commas are rejected.
  • Addresses: 0x + 40 hex characters, or a Tron T… address.
  • Invoice IDs: hex only, so a manipulated model can’t steer requests to other API paths.
  • Currency IDs and references: limited character sets and lengths.

get_payment_transactions flags token approvals that grant an unlimited allowance, so payers can lower them to the invoice amount in their wallet before signing.

Names, notes and payment links you give invoices are saved in a private file on your computer (~/.config/req-invoice-mcp/invoices.json, readable only by you). It’s never uploaded. Set REQUEST_NOTES_FILE to move it, or to off to keep nothing on disk.

Keys live in your MCP client config. They’re sent only to the Request Network API over HTTPS and never appear in tool output or error messages. A non-HTTPS API URL is refused.

Read-only tools are marked read-only, so MCP clients can auto-approve them while asking before anything that creates an invoice.

Every tool is covered by tests, including bad inputs, path-injection attempts, API failures and credential leaks. They run on every commit.

Open a GitHub issue. For anything sensitive, use GitHub’s private vulnerability reporting on the repository.