What is MUD?

What’s MUD?

MUD is a framework for ambitious Ethereum applications. It compresses the complexity of building EVM apps with a tightly integrated software stack.

MUD comes with:

  1. Store: An onchain database.
  2. World: An entry-point framework that brings standardized access-control, upgrades, and modules.
  3. Blazing fast development tools based on Foundry (opens in a new tab)
  4. Client-side data-stores that magically reflect onchain state.
  5. MODE: A Postgres database you can query with SQL that reflects your onchain state 1 to 1.

What MUD isn't

  • MUD is not a rollup or a chain: it’s a set of libraries and tools that work well together to build onchain applications.
  • MUD is not specific to Ethereum Mainnet: It works on any EVM-compatible chains. Polygon, Arbitrum, Optimism, Gnosis Chain, you name it.
  • MUD is not just for Autonomous Worlds and onchain games, although it has been a framework of choice in that community.
  • MUD doesn’t force a data model onto the developer (ie: not just ECS): anything you can do with flat Solidity mappings and arrays, you can do with MUD.
  • MUD doesn’t use an alternative Data Availability scheme. MUD apps have the same DA guarantee as regular Ethereum applications like ENS and Uniswap if they are deployed on Mainnet or on rollups.

The Main Ideas

1. All onchain state is saved in Store, the MUD onchain database

The way we currently deal with smart-contract state leads to a couple major problems:

  1. Coupling state and logic makes upgrading the logic very difficult. Solutions like proxies and diamonds make this situation more bearable, but it is far from a panacea.
  2. Solidity and Vyper hash the keys of mappings which makes the storage of smart contracts un-introspectable. Developers need to explain how their smart contract stores data to services like TheGraph in order to query them quickly.
  3. View functions expose a few ways to read data from smart-contracts onchain and off-chain, but if the developers don’t think about all the possible needs from 3rd party contracts ahead of time, it is almost impossible for other contracts to compose on their state.
  4. Introducing reasonable limitations (like flat structs and a limit on the amount of arrays a struct can contain) can make storage more efficient and cheaper for all users.
  5. There is no standard around storing data and emitting events to notify changes to off-chain applications. Each application has a bespoke event / view functions setup in order to bridge the onchain state to a client, which leads to a massive amount of spaghetti networking code for each frontend.

With MUD, you never use the Solidity compiler-driven data storage. No arrays, no mappings, no bool isPaused at the top of your contract definition. All state is saved and retrieved using Store: a gas efficient onchain database.

Store is like SQLite: it’s an embedded database hand-optimized in Yul. It has tables, with columns and rows. As an example, here is how you can implement a data-structure you'd write as mapping(uint => mapping(uint => address)) in regular non-MUD Solidity.

Vanilla Solidity:

mapping(address => mapping(address => uint256)) private allowances
// storing
allowances[address(0)][address(1)] = 10;
// getting
allowance = allowances[address(0)][address(1)]

MUD Store:

// table definition
Allowance: {
  keySchema: {
    from: "address",
    to: "address",
  schema: {
    amount: "uint256",
// storing
AllowanceTable.set(address(0), address(1), 10);
// getting
allowance = AllowanceTable.get(address(0), address(1));

Store supports singleton tables, which are equivalent to storing a single variable, like address contractOwner.

The recommended use of Store is through libraries generated using the MUD code-generation tool, but it can also be used directly with raw bytes.

[code with a schema → lib → writing and reading from Store]

Unlike Solidity where all data structures in storage need to be created at compile time, You can create tables at runtime: this is particularly useful to extend the amount of state a contract stores over time, or even to rename tables and columns using migrations. Store brings some of the sanity of SQL database to Ethereum, without trading-off on gas cost.

Store also allows you to register hooks on certain tables and rows to automatically create indexed views. For example: in case you’d want derived state like balanceOf in the ERC-721 spec, Store can recompute the corresponding row for an address every-time any piece of logic changes the owner of a token. It will decrease the balance of the sender and increase the balance of the receiver. There is no need to implement onTransferHooks : any piece of code that changes the owner column in the Token table will trigger a re-computation of the Balance table.

[code for that]

Store can either be used on the same contract running your business logic (tradeoff: more gas efficient, no upgradeability and contract size capped), or on another contract (tradeoff: slightly less gas efficient, the logic can be upgraded and span multiple contracts with optional per-table access control).

[diagram: store w/ logic, multiple logic contracts talking to Store]

Store uses a custom storage encoding called Tight-coder. It makes Store cheaper to use than regular Solidity for data with more than one dynamic field (eg: two arrays in a struct), and roughly similarly expensive for other data structures.

[two code stuff with raw Solidity and store, and gas cost commented]

2. Logic is stateless and partitioned across different contracts with custom permissions

Along with Store, MUD recommends the use of World: an entry-point kernel that takes care of mediating access to the Store from different different contracts.

World creates a Store at deployment time. Each table in the Store is registered under a namespace with a name, represented like a flat filesystem path. eg: /mudswap/BalanceTable. In this example, the namespace is mudswap and the name is BalanceTable.

Features — like the logic to transfer a token from one address to another — are added to the World via state-less pieces of logic we call Systems which are also registered under a namespace. eg: /mudswap/performSwapSystem

Systems are contracts, but they have no state. Instead, they read and write to the Store of the World. They can also be re-used across different Worlds if they are deployed on the same chain.

Systems can read any table, and they have default write access to tables registered under the same namespace they are in. In order to write to tables in different namespaces, explicit access needs to be granted by the address owning the corresponding namespace.

Systems exclusively using Store to read and write data makes it trivial to upgrade logic: one can simply redeploy the System and register it under its previous route. All other Systems that were referring on that route will now call the new contract, and write permissions for the System are conserved while writing to tables in the same namespace. Tables in different namespaces will need to re-approve write access when a System is upgraded.

This mediation of writes by the World — akin to system calls (opens in a new tab) in an operating system — allows Systems deployed and managed by different parties to co-exist around the same Store and thus share the same state.

[diagram with a World, one table, two systems, one can write to the table while the other one can’t]

The World determines which System can write to which table, and sending the corresponding write and read commands to the World is handled by the code-generated libraries.

All Systems are called through the World entry point, which does initial access-control check and forwards the call to the corresponding system with the original msg.sender added to the calldata in the EIP-2771 (opens in a new tab) fashion (_msgSender())

3. No indexers or subgraphs needed, and your frontend is magically synchronized!

When using Store (and by extension World), your onchain data is automatically introspectable, and any change is advertised via standard events.

These events and schemas are leveraged by MODE (opens in a new tab): a node, but for MUD*.* MODE turns your onchain state into a SQL database and keeps it up to date with millisecond latencies. You can then query MODE via the MUD QDSL (a querying language optimized for describing efficient materialized views), GraphQL, or regular ANSI SQL.

If you decide to go with the MUD querying DSL, your app can subscribe to a subset of the Store via a simple yet flexible querying language and keep that subset of the Store up to date as onchain transactions change Store and the result set.

[diagram with a few tables, a DSL query, a client that updates its frontend, and an onchain tx that percolates to changing the UI]

This means you don’t need to write Subgraphs or Indexers: you just need to point a MODE to your Store!

A major pain of frontends and clients is reflecting onchain state without spaghetti code. Some apps resort to polling, with various tricks like Multicall or Subgraphs in order to decrease RPC costs.

With MUD, you define a list of queries on your Store state (the simplest being * which asks for all the data). The MUD client then either connects to a MODE or an Ethereum JSON-RPC in case none is available and magically keeps the Store data up to date on your frontend. We currently have libraries for Javascript, with more on the way. If you use Typescript, all your tables will be typed automatically.

If you use React, MUD provides React hooks to bind the Store state to your components and automatically re-render them when the onchain data changes.

[diagram with a balance table → a frontend with a button to transfer and an addressed prefilled → a onchain tx to a system → balance table updates → frontend update]