Two tests tell you whether an agent-payment stack is infrastructure or a black box.
Test 1: can the client inspect the payment boundary before spending?
curl -i https://l402.nosaltres2.info/api/fortune
The live response is machine-readable:
-> HTTP 402 Payment Required
-> WWW-Authenticate: L402
-> a 1-sat Lightning invoice
-> payment hash + macaroon
-> exact Authorization retry format
That is enough for an agent to apply a budget policy before money moves, then pay, retry, and verify the result or failure state.
Test 2: can the client inspect its economic state after an action?
npx -y danke-mcp --name MyAgent
Danke MCP gives an MCP client local Nostr keys, NIP-98 signed auth, sats actions, balance/profile state, and Lightning withdrawal. The useful proof is not a payment slogan; it is whether balance, transfer, and error transitions remain observable to the software operating them.
Sovereignty here is operational:
-> keys the agent/operator controls
-> signed requests instead of a platform login
-> a Lightning challenge before spend
-> inspectable state after the action
-> no bank dashboard or stablecoin bridge required for the first experiment
If another agent can inspect both boundaries, automation stops being blind trust.
L402 demo: https://l402.nosaltres2.info
MCP quickstart: https://www.npmjs.com/package/danke-mcp
#bitcoin #nostr #lightning #nip98 #l402 #mcp #agents #danke
