# Shield47

> Privacy infrastructure for the agent economy. Shield47 shields agent transactions into a unified privacy pool and unshields to a clean wallet with zero linkage between the two. Non-custodial, live on Base, verified by zero-knowledge proofs, and agent-native (MCP server + OpenClaw SKILL.md).

Shield47 is privacy infrastructure for AI agents transacting on-chain. On a public blockchain, anyone can read an agent's entire on-chain life: balances, counterparties, swaps, recurring payments, positions, and labels. Shield47 lets an agent shield funds into one shared anonymity set and later unshield to a fresh wallet, with zero-knowledge proofs severing the link between deposit and withdrawal.

This file is intended for AI agents and crawlers. It states the live-vs-roadmap status honestly: roadmap items are never presented as shipped.

## What is live today

- Link-breaking shielded transfers, running on Base.
- Proof system: Groth16 zero-knowledge proofs. The live stack uses a trusted setup.
- Independent security audit: complete.
- Unified privacy pool: one anonymity set; every shield deepens the same pool.
- Non-custodial: the protocol never takes custody. You hold the private note; only that note can unshield.
- Gasless withdrawals: a relayer pays gas via x402, so a fresh wallet needs no prior funding.
- Compliance-capable: KYT/KYP screening keeps flows auditable for institutions; Shield47 is privacy infrastructure, not a sanctions-evasion tool, and screens against illicit flows.
- Agent-native: a native MCP server is operational; works with Claude, Cursor, and ACP.

## On the roadmap (NOT yet deployed)

- Post-quantum stack based on WHIR (hash-based proofs), to stay secure even against a future quantum computer that could break today's elliptic-curve cryptography.
- WHIR proof circuit: no trusted setup.
- Formally verified in Lean: in progress.
- On-chain verification: approximately 3M gas.
- Deployment: not yet deployed anywhere. Status approximately 95% complete.

## How it works

1. Shield: your agent deposits funds into the unified privacy pool and receives a private note.
2. Blend: zero-knowledge proofs sever the connection to your deposit inside one shared anonymity set.
3. Unshield: funds arrive at a fresh wallet with no traceable link to the source; gasless via x402.

## Positioning

Most privacy approaches hide amounts but leave the payer-to-payee link visible, so the transaction graph is still reconstructable. Isolated privacy chains are walled off from DeFi liquidity and from on-chain agents. Shield47 breaks the link, on Base, with full DeFi and agent reach, and stays compliance-capable for institutions.

## Token

- $S47: fee share, fee discounts, and governance. Fee percentage is not yet finalized.

## For developers and agents

- Native MCP server: drop into Claude, Cursor, and ACP.
- OpenClaw SKILL.md: lets agents shield and unshield with no frontend required.
- Shield and unshield from a few lines of agent code (package: @shield47/mcp).

## Links

- Site: https://shield47.ai/
- Sitemap: https://shield47.ai/sitemap.xml

## Honesty note

Several on-page figures are marked illustrative and are placeholders to be replaced with verified values before launch. Live capabilities and roadmap capabilities are kept strictly separate: do not attribute roadmap features (post-quantum, WHIR, Lean verification, no-trusted-setup) to the currently live system, which uses Groth16 with a trusted setup.