---
title: "Radix for AI Agents"
url: "https://radix.wiki/contents/tech/core-concepts/radix-for-ai-agents"
updated: 2026-08-19
last_verified: 2026-08-18
license: CC-BY-4.0
license_url: "https://creativecommons.org/licenses/by/4.0/"
version: "1.4.1"
---

# Radix for AI Agents

|  |  |
| --- | --- |
| **Category** | Use Case |
| **Key Properties** | Readable transactions, no approvals, atomic ops, linear scale |
| **Related** | [Radix Context for AI](/developers/ai-agents/radix-context) |
| **Agent Interface** | [Model Context Protocol](https://modelcontextprotocol.io) servers |

## Overview

As AI agents increasingly manage financial operations autonomously, the underlying ledger's architecture becomes critical. Radix provides four properties that make it uniquely suited for **autonomous AI agent operations**:

### 1. Machine-Readable Transactions

[Transaction manifests](/contents/tech/core-protocols/transaction-manifests) are human AND machine readable. An AI agent can parse, verify, and generate manifests – understanding exactly what a transaction will do before signing. On EVM chains, [blind signing](/contents/tech/core-concepts/blind-signing-and-transaction-manifests) means agents sign hex data they can't interpret, creating catastrophic risk.

### 2. No Approval Exploits

The [ERC-20](https://eips.ethereum.org/EIPS/eip-20) `approve()` pattern requires granting contracts access to your entire balance. For an AI agent, this means trusting every contract it interacts with. On Radix, [native assets](/contents/tech/core-concepts/native-assets-vs-token-approvals) move directly – no approval surface to exploit.

### 3. Atomic Composability

Agents can compose complex multi-step strategies in single [atomic transactions](/contents/tech/core-concepts/atomic-composability). No partial failures, no stuck funds, no error recovery needed.

### 4. Linear Scalability

When millions of agents transact simultaneously, gas auctions don't scale. [Xi'an](/contents/tech/releases/radix-mainnet-xian) delivers linear throughput – more nodes means more capacity.

## Connecting an agent

The properties above describe the ledger. What an agent actually talks to is a server, and on Radix that server is almost always an [Model Context Protocol](https://modelcontextprotocol.io) (MCP) endpoint — the standard by which a model client discovers a set of tools and calls them. Six are in public use, at different layers of the stack:

- **[RadixScan](/ecosystem/radixscan)** – a keyless MCP server at `mcp.ai.radixscan.io` that reads ledger state and builds, validates and simulates [transaction manifests](/contents/tech/core-protocols/transaction-manifests), then hands the finished manifest to the user's wallet for signature. It never holds a key.
- **[AgentWallet](/developers/ai-agents/agent-wallet-ai)** – exposes the same wallet, swap and lending operations through both a [LangGraph](https://www.langchain.com/langgraph) toolset and an MCP server, so an agent can drive a funded account on mainnet or [Stokenet](/contents/tech/releases/stokenet).
- **[`radixdlt-connector-mcp`](/developers/infrastructure/radixdlt-rust-sdk)** – a local MCP server from the GenkiPool Rust SDK that pairs a Radix Wallet over [Radix Connect](/contents/tech/core-protocols/radix-connect) and requests signatures on the user's own machine. It is installed from the repository rather than [crates.io](https://crates.io), where none of the SDK's crates are published.
- **[Igentix](/developers/ai-agents/igentix)** – a Stokenet demonstration at `demo.igentix.app/mcp` where an agent discovers a priced service, pays for it under [x402](/developers/ai-agents/ai-agents-and-x402), and receives the result with no gas token and no wallet pop-up.
- **[Radix Community](/ecosystem/genkipool)** – a hosted, keyless server at `radix-community.genkipool.com/api/mcp` and the largest of the set: 34 tools, every one of them annotated read-only. It answers from three layers at once – the site’s curated documentation, the ledger (entities, balances, transactions, validators, NFT and component state), and its browser developer console, where the manifest-building tools return a [manifest](/contents/tech/core-protocols/transaction-manifests) the user signs from the console with their own wallet. Same operator as `radixdlt-connector-mcp` above, and a different thing: that one runs on the developer’s machine and pairs a wallet over [Radix Connect](/contents/tech/core-protocols/radix-connect); this one is a public endpoint that holds nothing.
- **This wiki** – `radix.wiki/api/mcp` serves the encyclopedia itself over the same protocol: search, read and list tools are open and unauthenticated, and an agent that signs a [ROLA](/developers/frontend/03-rola-authentication) challenge with its own key can also create and edit pages. The full contract is at [AGENTS.md](/AGENTS.md), and any page URL with `.md` appended returns that page as markdown.

The pattern RadixScan, AgentWallet, `radixdlt-connector-mcp` and Radix Community share is the one that makes the [manifest](/contents/tech/core-protocols/transaction-manifests) matter: the server composes the transaction, the human's wallet signs it. An agent can be given the ability to propose any operation without being given custody of anything.

## Developer Resources

The [radix-context](/developers/ai-agents/radix-context) project provides 19 curated files that teach AI coding agents to build on Radix, covering [Scrypto](/contents/tech/core-protocols/scrypto-programming-language), [Gateway API](https://docs.radixdlt.com/docs/network-gateway), [transaction manifests](/contents/tech/core-protocols/transaction-manifests), and wallet integration.

## External Links

- [radix-context on GitHub](https://github.com/xstelea/radix-context)
- [Blind Signing and Transaction Manifests](/contents/tech/core-concepts/blind-signing-and-transaction-manifests)