Summary: Lightning Agent Tools packages the Lightning Labs view of agent commerce on Lightning: node access, remote-signer isolation, macaroon scoping, L402 buying, paid-endpoint hosting, and read-only MCP access. Useful because it bundles the whole operating stack in one place, not because it invents a new rail.
Lets an agent run or connect to a Lightning node, pay for L402-gated resources, and optionally host paid endpoints behind an L402-aware reverse proxy
Uses a default remote-signer architecture so the agent machine can operate a watch-only node while private keys stay on a separate signer machine
Provides scoped macaroon roles such as pay-only, invoice-only, read-only, channel-admin, and signer-only so agent permissions can be limited to a specific operational task
Exposes a read-only MCP server that connects through Lightning Node Connect (LNC) using a 10-word pairing phrase and a mailbox relay instead of direct inbound node access
Supports end-to-end buyer and seller workflows for agent commerce by tying together node management, spending credentials, L402 payment handling, and paid API hosting
Key claims:
The repository README says the toolkit consists of seven composable skills and an MCP server that together let agents run a Lightning node, pay for resources on the web using L402, host their own paid API endpoints, manage scoped credentials, and query node state through MCP
The README says the security model defaults to a remote signer architecture that keeps private keys on a separate machine away from the agent runtime, which is a major clue that the project is as much about operational security as it is about payments
The security doc defines three access tiers and explicitly recommends the watch-only plus remote-signer setup as the default production posture, while the agent machine lacks the wallet seed and private keys
The lnget documentation in the README and launch post says the CLI automatically handles 402 responses by paying the embedded Lightning invoice, caching the token, and retrying the request, which makes it a true machine-payment client rather than only a manual helper
The MCP docs say the lightning-mcp-server exposes 18 read-only tools over Lightning Node Connect and stores no credentials on disk, using an ephemeral ECDSA keypair and encrypted WebSocket tunnel through a mailbox relay
The MCP docs also make clear that the read-only MCP-LNC path is intentionally separate from direct gRPC control, reinforcing that observation and action are split into different trust and permission models
The Lightning Labs launch post positions the toolkit as open-source infrastructure for agents that need identity-free, programmatic micropayments, and it specifically ties together lnget, Aperture, remote-signer security, and L402 as one coherent commerce stack
Whitepaper: No canonical standalone Lightning Agent Tools whitepaper or litepaper surfaced in this pass. The clearest source of truth was the GitHub repo, raw README, dedicated docs for security and MCP setup, and the Lightning Labs launch post; see ../whitepapers/lightning-agent-tools-primary-sources-2026-05-03.md.