Agent API

BillWatch is built to be bought by AI agents on behalf of the humans they support. Two ways to connect: a REST API and an MCP server. Both are read-mostly by design — approving or dismissing a finding always requires the human's explicit go-ahead, relayed by their agent.

REST API

Base URL: https://getbillwatchdog.com (placeholder — domain attaches after launch)

Auth: Authorization: Bearer <api-key>. Keys are issued in-app; only SHA-256 hashes are stored.

Method	Endpoint	What it does
GET	`/v1/findings`	List detected findings (price hikes, forgotten subs, duplicates)
GET	`/v1/summary`	Counts, estimated monthly savings, commissions earned
POST	`/v1/findings/{id}/approve`	Human-approved: bank the estimated saving
POST	`/v1/findings/{id}/dismiss`	Human-dismissed: stop flagging this finding

MCP server

For agents that speak the Model Context Protocol. Tools:

list_findings — the current findings for the connected human
get_summary — savings totals and counts
approve_finding / dismiss_finding — record the human's decision

Manifest: /.well-known/agent.json · Machine-readable index: /llms.txt

Ground rules for agents

Never act on a finding without the human's explicit approval. Present the finding, the saving, and the options.
BillWatch is a spotter, not a handler: it cannot pay, cancel, or switch anything by itself.
Nothing here is financial advice. Surface facts; let the human decide.
Sandbox mode is clearly labelled. Real bank connections arrive only after independent legal review.