DEV Community

Baris Sozen
Baris Sozen

Posted on

A week in intent-based trading: Ghost Auction, live quotes, and the path to agent-native DeFi

A short Sunday recap from the Hashlock Markets team: what shipped this week, what it changes for traders and market makers, and what the next sprint is focused on.

If you're new here: Hashlock Markets is an intent-based trading protocol. You declare what you want to trade and on what terms. Market makers respond with sealed-bid quotes. The winning quote settles via a Hash Time-Locked Contract (HTLC) — atomic, cross-chain, no escrow, no information leakage. The whole stack is also accessible via the Model Context Protocol, so an AI agent (Claude, GPT, anything that speaks MCP) can run the flow end-to-end.

1. Ghost Auction: sealed-bid pricing, with a clearer name

The protocol's RFQ pricing has always been sealed-bid: market makers cannot see each other's quotes; only the trader can. We renamed the user-facing surface from blind auction to Ghost Auction. The mechanism didn't change — the framing did. Sealed bids deliver three properties that matter in DeFi:

  • No information leakage. A market maker's quote is private to the trader.
  • No front-running. No public mempool order to sandwich.
  • Better fills. Each MM has to put in their best independent price; no copying off the order book.

2. Live quote feed

When a market maker submits a price on your RFQ, you see it the moment it arrives — anywhere in the app. Toast notification, optional sound, click-through to the trade detail. Sealed-bid doesn't need to mean slow. The trader sees every incoming quote in real time, and the auction window stays short.

3. End-to-end idempotency on the RFQ path

Idempotency keys now cover the full RFQ lifecycle: creating an RFQ, submitting a quote, and accepting a winning quote. AI agents retry — that's a feature of how they work — and the protocol now treats retried calls as a no-op when they arrive with the same key. Concretely, this matters most for AI agents using create_rfq, respond_rfq, and the acceptance step: you can rerun the whole flow safely without double-creates.

4. Why this matters for AI-agent trading

Hashlock Markets exposes six MCP tools that map onto the trading lifecycle:

Tool What it does
create_rfq Open a sealed-bid RFQ
respond_rfq Submit a quote as a market maker
create_htlc Lock funds in the atomic settlement contract
withdraw_htlc Claim an HTLC by revealing the preimage
refund_htlc Refund after timelock expiry
get_htlc Query lifecycle state

Two ways to connect: stdio (npx -y @hashlock-tech/mcp) or Streamable HTTP at https://hashlock.markets/mcp. SIWE (Sign-In with Ethereum) handles auth; the agent never needs to handle a password.

5. What's next

A few things on deck for the coming weeks:

  • More chains. Solana and Arbitrum are next on the HTLC integration roadmap, joining Ethereum, Bitcoin, and Sui.
  • Reference market-maker template. A working example MM stack that other teams can fork — pricer, RFQ subscriber, quote handler, settlement.
  • Agent-mode polish. Tighter error semantics, richer get_htlc state, and better discoverability via llms.txt.

If you're an agent developer or a market maker, the canonical repo is at github.com/Hashlock-Tech/hashlock-mcp. Drop the MCP server into your client and you can place a real cross-chain trade with create_rfq in a few minutes.

— Built at Hashlock Markets

Top comments (0)