Agentic Vault MCP Server
Exposes an EVM wallet to AI agents via MCP, using AWS KMS for HSM-backed signing and a deny-by-default policy engine for secure DeFi transaction approvals.
README
Agentic Vault
<p align="center"> <img src="https://raw.githubusercontent.com/agenticvault/agentic-vault/main/docs/img/cover.gif" alt="Agentic Vault" width="100%"> </p>
English | 繁體中文 | 简体中文 | 日本語 | 한국어
Server-side EVM signing with AWS KMS and built-in DeFi protocol awareness. Expose your wallet to AI agents via MCP, CLI, or OpenClaw with deny-by-default policy enforcement and full audit logging.
Why Agentic Vault
AI agents need to sign blockchain transactions, but giving them private keys is dangerous. Agentic Vault solves this by keeping keys in AWS KMS (HSM) and providing a policy engine that constrains what agents can sign. The agent sees high-level tools (sign_swap, sign_permit); the private key never leaves the hardware.
Features
- HSM-backed signing -- private keys never leave AWS KMS; only digests are sent for signing
- DeFi protocol awareness -- calldata decoding for ERC-20, Uniswap V3, and Aave V3 with protocol-specific policy rules
- Deny-by-default policy engine -- chain, contract, selector, amount, deadline, and protocol-level constraints
- Multiple interfaces -- use as a TypeScript library, CLI, MCP server, or OpenClaw plugin
- Audit logging -- structured JSON audit trail for every signing operation (approved, denied, errored)
- EVM-native -- built on viem with full EIP-712 typed data support
Quick Start
Requires Node.js 22+.
npm install @agenticvault/agentic-vault
import { createSigningProvider, EvmSignerAdapter } from '@agenticvault/agentic-vault';
const provider = createSigningProvider({
provider: 'aws-kms',
keyId: 'arn:aws:kms:us-east-1:123456789:key/your-key-id',
region: 'us-east-1',
});
const signer = new EvmSignerAdapter(provider);
const address = await signer.getAddress();
For a no-AWS quick test, use dry-run mode (decode + policy check without signing):
npx agentic-vault dry-run --chain-id 1 --to 0xa0b869... --data 0x095ea7b3...
See AWS KMS Setup for key creation and IAM policy.
Interfaces
| Interface | Use Case | AWS Required |
|---|---|---|
| TypeScript Library | Embed signing in your app | Yes |
| CLI | Command-line signing + dry-run | Partial |
| MCP Server | Expose wallet to AI agents (Claude, etc.) | Yes |
| OpenClaw Plugin | Use as OpenClaw agent tool | Yes |
See Interfaces Guide for usage examples and configuration.
Supported Protocols
| Protocol | Actions | Decoder | Policy Evaluator |
|---|---|---|---|
| ERC-20 | approve, transfer |
Yes | Yes (allowance cap, spender allowlist) |
| Uniswap V3 | exactInputSingle |
Yes | Yes (token pair, slippage, recipient) |
| Aave V3 | supply, borrow, repay, withdraw |
Yes | Yes (asset allowlist, interest rate mode) |
Unknown calldata is rejected by default (fail-closed). The dispatcher uses 2-stage resolution: contract address first, then selector-based fallback (e.g., ERC-20). Raw signing tools (sign_transaction, sign_typed_data) bypass protocol decoding and are disabled unless --unsafe-raw-sign is explicitly enabled.
Configuration
The policy engine uses a JSON configuration file. Without a policy file, all policy-guarded signing operations are denied (deny-by-default).
See Policy Reference for the full schema and examples, or start with policy.example.json.
Security Model
Trust Boundary
AI Agent (Claude / MCP Client / OpenClaw)
|
| MCP Protocol / OpenClaw Plugin API
v
+------------------------------------+
| Agentic Vault |
| +-----------+ +--------+ +-----+ |
| | Protocol | | Policy | | Audit| |
| | Dispatcher| | Engine | | Sink | |
| +-----------+ +--------+ +-----+ |
| | |
| +--------------------+ |
| | EvmSignerAdapter | |
| +--------------------+ |
+-----------|------------------------+
| digest only
v
+------------------------------------+
| AWS KMS (HSM) |
| Private key never leaves |
+------------------------------------+
Key Principles
| Principle | Description |
|---|---|
| Key isolation | Private keys remain in the HSM; only 32-byte digests are sent for signing |
| Deny by default | Policy engine rejects all requests unless explicitly allowed |
| Fail-closed | Unknown calldata is rejected by default; raw signing tools are disabled unless --unsafe-raw-sign is enabled |
| Audit trail | Every operation is logged as structured JSON to stderr with caller tag |
| Minimal surface | Raw signing tools (sign_transaction, sign_typed_data) are disabled by default |
Claude Code Plugin
4 skills interact with the wallet through MCP tools only -- they never access keys directly.
| Skill | Description |
|---|---|
sign-swap |
Orchestrate a swap signing operation |
sign-permit |
Orchestrate an EIP-2612 permit signing |
check-wallet |
Check wallet address and health status |
audit-log |
Query the audit log |
Package Exports
| Subpath | Contents | MCP dependency |
|---|---|---|
@agenticvault/agentic-vault |
Core signing (SigningProvider, EvmSignerAdapter, factory) | No |
@agenticvault/agentic-vault/protocols |
Protocol decoders, dispatcher, PolicyEngine V2, workflows | No |
@agenticvault/agentic-vault/agentic |
MCP server, audit logger | Yes |
Documentation
| Document | Description |
|---|---|
| Interfaces Guide | TypeScript, CLI, MCP, and OpenClaw usage |
| Policy Reference | Policy JSON schema, fields, and examples |
| AWS KMS Setup | Key creation, IAM policy, authentication |
| OpenClaw Plugin | OpenClaw plugin package and configuration |
| Architecture Decisions | ADRs for key design choices |
| Contributing | Development workflow and guidelines |
Roadmap
- Additional signing providers (GCP KMS, HashiCorp Vault)
- More protocol decoders (Curve, Compound V3)
- Multi-signature support
- Remote MCP server mode (HTTP transport with OAuth 2.1)
Contributing
See CONTRIBUTING.md for development workflow, branch naming, and commit conventions.
This project follows the Contributor Covenant Code of Conduct.
License
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.