惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

V
V2EX
宝玉的分享
宝玉的分享
Jina AI
Jina AI
IT之家
IT之家
博客园 - Franky
MyScale Blog
MyScale Blog
Y
Y Combinator Blog
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
I
InfoQ
雷峰网
雷峰网
WordPress大学
WordPress大学
Microsoft Security Blog
Microsoft Security Blog
Google DeepMind News
Google DeepMind News
美团技术团队
S
SegmentFault 最新的问题
罗磊的独立博客
博客园 - 聂微东
大猫的无限游戏
大猫的无限游戏
H
Help Net Security
D
Docker
博客园 - 司徒正美
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
阮一峰的网络日志
阮一峰的网络日志
M
MIT News - Artificial intelligence

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant
Building Multi-Chain Support for Ironclaw
Den · 2026-05-01 · via DEV Community

"I'd like to send 0.005 ETH to my friend 0x1234... on Sepolia. My NEAR address is xxxx.testnet and the derivation path is the string ethereum-0."

That was the prompt I typed into Ironclaw last week. The agent derived a Sepolia address it had never seen before, looked up gas prices and a nonce, built the transaction, asked a contract on NEAR to sign it, and broadcasted the result. Just 10 seconds later it handed me an Etherscan link.

That single transfer was the end of a project to get Ironclaw signing transactions on any chain. Here's how we built it, in two parts.

What's Ironclaw?

For readers who haven't met it - IronClaw is OpenClaw inspired implementation in Rust focused on privacy and security. It runs on your own machine, talks to whichever language model you point it at, and remembers what you tell it across sessions.

Teaching the agent to use a NEAR account

The plan was multi-chain support from day one: we wanted Ironclaw to send a transaction on any chain its user cared about. But the route to "any chain" ran through NEAR specifically, because of a contract on NEAR called the chain-signatures MPC contract — more on that in a moment. To use that contract, the agent would first need to be able to send transactions from the user's NEAR account.

So the first milestone was narrower - full NEAR-account control. Get that working, and we'd have everything we needed to start on the rest.

The first attempt was the obvious one - build a NEAR-signing tool. Compile it to WebAssembly, plug it into Ironclaw, let the agent call it. We got partway through before hitting the wall. Ironclaw's tool sandbox is intentionally restrictive about secrets, a tool can ask whether a credential exists, but never read it. Private keys included. A signer-as-a-tool was a non-starter.

The second attempt sat at a different layer. Instead of building a signer ourselves, we wrote a skill that taught the agent how to drive near-cli-rs, the user's existing command-line tool for NEAR, from the shell. Shell access in Ironclaw is opt-in and off by default; you launch the agent with ALLOW_LOCAL_TOOLS=true ironclaw to enable it. The skill then describes the surface of near-cli-rs to the agent, which subcommand maps to which intent, what arguments to compose, how to read the output back.

That worked. The agent could now turn a sentence like "send 0.5 NEAR to bob.testnet" into the right near-cli-rs invocation and run it. Milestone reached — and although NEAR-account control was useful in its own right, its real importance was as the foundation for what came next.

Going multi-chain via NEAR MPC

What is MPC?

NEAR runs an MPC chain-signatures contract at v1.signer-prod.testnet, hand it a payload and a (account, path) pair, and it returns a signature, produced by a key that's derived deterministically from those two inputs. Same NEAR account, infinite addresses on every chain you know how to derive for Ethereum, Bitcoin, Solana, all from the one NEAR identity.

The architecture

A small Rust WASM tool (near-tool) with a set of deterministic actions, plus a skill (MPC_SKILL.md) that teaches the agent how to use them, plus the near-cli-rs shell command plumbing from Step One. The tool handles the chain-specific encoding. The skill orchestrates the sequence. near-cli-rs makes the actual MPC contract call.

When I sent the prompt from the beginning, the agent walked through roughly this sequence:

NEAR account + path
        │
        ▼
1.  get_derived_pubkey                ──►  Sepolia address
        │                                  (and BTC/Solana too)
        ▼
2.  evm_parse_value                   ──►  amount in wei (hex)
        │
        ▼
3.  evm_get_nonce                     ──►  pending nonce
        │
        ▼
4.  evm_get_gas_price                 ──►  base gas price
        │
        ▼
5.  evm_get_priority_fee_wei_per_gas  ──►  priority fee
        │                                  + max fee per gas
        ▼
6.  evm_estimate_gas                  ──►  gas limit (20% buffer)
        │
        ▼
7.  evm_build_transfer_mpc_payload    ──►  unsigned tx
        │                                  + sighash payload
        ▼
8.  near-cli-rs sign payload via MPC  ──►  (big_r, s, recovery_id)
        │
        ▼
9.  evm_attach_mpc_signature_to_tx    ──►  signed tx + tx_hash
        │
        ▼
10. evm_send_signed_tx                ──►  Etherscan link

Enter fullscreen mode Exit fullscreen mode

A few build choices worth pointing at.

The agent narrates every step. The nonce, the gas estimate, the unsigned transaction, the raw signature — they all show up in the chat. It was tempting to bundle the prelude into one big "build me a transaction" call, but legible failure modes were worth more than the cleaner-looking flow.

That's also why steps 2 through 6 are separate actions instead of one. When something breaks, a flaky RPC endpoint, an underestimated gas limit, the agent knows exactly which piece broke, can show the user a sensible cost before committing, and can retry just the part that needs retrying.

What's next

Bitcoin and Solana already work end-to-end through the same arc, but their flows are coarser than EVM's — fewer intermediate steps, fewer natural seams to surface errors. The next push is to bring them up to the same level of granularity: small actions for fee estimation, UTXO selection, blockhash fetching, and so on.

Additional resources