Skip to main content

​
MCP

Fortytwo MCP is an orchestration layer that allows any MCP-compatible client to access Fortytwo Network directly.It implements the Model Context Protocol — a standard that exposes available tools to applications, AI models, and agents enabling them to interact with external services.
Yes. MCP is a protocol that exposes Fortytwo tools to applications, models, or agents so that they can easily access Fortytwo.
Yes. MCP works without an API key. Instead, x402Escrow protocol will serve as your authorization and payment method by utilizing your web3 wallet.
x402Escrow enables low-cost payments for AI inference while allowing the user to remain in control of their wallet and expenses.x402Escrow is a special variation of x402 protocol that is designed by Fortytwo specifically to enable usage-based, pay-per-token billing for AI services, charging only for the tokens used. It works in two steps:
  1. Estimated USDC amount is reserved in an escrow contract.
  2. Inference result is provided to the user, the amount for that result is charged, and the remaining funds are released back to the user.
The key difference between x402 and x402Escrow:
  • x402 protocol relies on fixed price endpoints where costs are known before the operation.
  • x402Escrow is specifically built to handle AI inference usage cases, where the costs are unknown at request time and only become known when inference is completed.
Learn more:
x402Escrow requires at least 2 USDC on Base or Monad networks. This should be enough to cover multiple requests within a session.
USDC payments are accepted on:
  • Base: eip155:8453
  • Monad: eip155:143
The contract is deployed on Monad and Base networks:
Monad: 0x9562f50f73d8eE22276F13A18D051456d8D137a0
Base: 0x9562f50f73d8eE22276F13A18D051456d8D137a0
The server responds with 410 Gone or 402 Payment Required. Sign a new payment and start a new session. See MCP Integration -> Step 5 for details.
A fallback refund is available on-chain via refundAfterTimeout(escrowId) after the escrow’s refundAt deadline (~90 minutes). See MCP Integration -> Fund Release for details, or contact support.

​
API

Yes. Read ‘Integration with Chat Apps’ to learn how.
This page currently covers Jan and will be expanded with other apps over time.

​
P2P Network

To be able to see the FOR devnet token, you’ll need a wallet that allows importing custom tokens, like MetaMask .
  1. In your wallet, choose option Import Tokens.
  2. Select a network: Monad Testnet.
  3. Define the token contract address: 0xf6B888f442277F01294F94D555608A2E8Bc86430
  4. Give it a name: FOR(DEVNET)
After this, it should appear in your wallet. You might be required to select the Monad Testnet network as a category first. FOR(DEVNET) token that you just added will appear in this network’s list.

​
P2P Inference Nodes

​
Devnet Program

To receive an activation code, fill out our participation form at https://tally.so/r/wQzVQk . You will receive a single-use activation code in an e-mail. See the ‘Participate in the Devnet Program’ page for details.
Yes. Just follow this ‘Quick Start’ guide. With no activation code, you’ll need to manually monitor and top-up MON and FOR balance for your node to work, with 2 MON and 100 FOR required to start a node. Your node will not automatically receive MON refills with no activation code.
No. Activation codes are one-time use only.
It depends on how you participate in the network:

Case A: Your node receives automated gas compensation for participation.

Do not transfer MON out of your wallet in this case. This will terminate the compensation cycle on your node and you will have to fund your participation by yourself. You will not be able to return to the compensation program again. Your participation in the Devnet program will continue as usual as long as there is enough MON to pay for gas.

Case B: You cover your node's gas consumption manually.

You are free to refill and transfer MON out of your wallet as needed.
You can read more on ‘Holding MON in Your Wallet’ page.
If you used an activation code, your wallet is usually refilled within 24 hours. If it’s been longer, message us on Discord . Reminder: auto-refill works only for wallets tied to activation codes.
You can get MON testnet tokens via the official Monad’s Faucet .
Wallet requirements for claim eligibility:
  • The wallet must have at least 10 MON on mainnet
  • One wallet can claim tokens once every 6 hours

Situation A: Self-funded nodes

If you don’t have enough testnet tokens:
  1. Create multiple wallets. Amount depends on your spending.
  2. Fund each wallet with ≥ 10 MON on mainnet.
  3. Claim testnet tokens from the faucet, once every 6 hours for each wallet.
  4. Send all testnet tokens to the wallet running your node.

Situation B: Fortytwo-funded nodes fallback

If your node is participating in Fortytwo MON refund program, but:
  • testnet tokens are depleted,
  • and auto top-up hasn’t arrived yet,
you can temporarily use the method from Situation A to get testnet tokens and avoid node downtime. Auto top-up will continue working as usual.

​
Running the Inference Nodes

If you haven’t used it yet, ping our support team on Discord . If you’ve already activated it, just use your private key or recovery phrase to log in instead.
No. Only one node per wallet is supported. You will face issues or node inactivity if you attempt to use the same wallet on several nodes simultaneously.
It depends on the following conditions:
  • Fortytwo P2P CLI and Fortytwo P2P App
    No. To avoid issues with your node, only run one node instance per machine. Note that using Fortytwo P2P CLI and Fortytwo P2P App counts as two node instances.
  • Fortytwo P2P Container
    Yes. It, however, will only make sense if you have a multi-GPU setup and want to run one node per particular GPU(s). In this case several Fortytwo P2P Container nodes can be launched on one device.
Yes. Use Fortytwo P2P Container. It allows to assign a particular node to a particular GPU.
Currently, Fortytwo P2P CLI and Fortytwo P2P App do not support this setup.
Not at this time. We support only GPU-based nodes due to performance requirements.
It depends on a few factors like machine specs, internet connection speed, and current system load. Try a few available models to see what works best on your setup.For more, read ‘Picking a Model for Self-Hosted Node’.
  • Your node spends 1 FOR to enter each inference round (shown as -1 FOR in explorers).
  • +X FOR rewards are granted when your node secures a winning place in an inference round. You can see which winning place your node has secured in the transaction details. Best answer yields the biggest reward.
  • Your node can enter a round and not secure a winning place, which will result in no rewards. The most common causes are:
    • The node took too long to provide an answer. This often happens when a model is too large for the system it is running on, or a connectivity issue has interrupted your connection to the network.
    • Your node generated a weak answer and therefore has not secured a winning place in an inference round.
This may be due to connection issues or low request volume in the network. Try restarting the node or switching to a different model. Sometimes it just needs a bit of time. If it’s been over 24 hours with no FOR earnings, reach out to us on Discord .
Yes, but an Nvidia GPU is required. CPU-only machines are not supported at this time. Read ‘Hardware Requirements’ for your system.
Export your private key and Secret Recovery Phrase from your virtual machine. If your VPS is lost, you will lose access to your node if your credentials are not exported.
We do not recommend using custom scripts unless you are fully confident in their origin and completely understand their functionality.
We cannot guarantee your privacy and security in this case or support should any issues arise.

​
P2P Applications

Yes. The nodes support models in GGUF format.
Note that not all GGUF models are immediately supported.
Use the search bar in the model manager to find models from remote repositories or import a locally downloaded GGUF model file by pressing the Import button in the model manager.
There are usually two scenarios:
  1. Free up resources taken by other applications. Closing ‘expensive’ processing applications such as 3D editors, video editors and video games might help.
    Then launch the node again and select your model. If you are manually controlling the KV Cache size, you can now reopen the applications that you closed.
  2. Try selecting a model that is about 2GB smaller than your total memory size. Bear in mind that, in addition to your model, memory will also be used by the KV Cache and the default processes of your OS.
See our ‘Picking a Model for Self-Hosted Node’ article for details.
Restarting your other applications should help.When you launch a resource-intensive application while your node is running, the application may only detect the currently available resources. Therefore, when you stop the node, your other application is unable to detect any additional resources that got freed up and may continue to experience issues until it is restarted.
See: ‘Inference Support’.
See: ‘Inference Support’.

​
P2P Relay Nodes

Yes. To run a Fortytwo P2P Relay Node, fill out our participation form at https://tally.so/r/9qN9Ap .You will be able to connect to the network as a Relay node once you receive an invitational e-mail.Being a Relay Noderunner requires additional knowledge and has specific requirements. See the ‘Relay Node Requirements’ for more details.
You receive FOR tokens for being online and providing high-quality, reliable routing.
This may be due to connection issues on your Relay node. Check your system’s internet connection. If it’s been over 48 hours with no FOR earnings, reach out to us on Discord .
Yes. For that you need to set unique FT_ACCOUNT_PRIVATE_KEY, FT_NODE_LISTENER_PORT, and FT_RPC_SERVICE_PORT values in the .env file. Refer to ‘Quick Start’ for more information on Relay node setup.
See: ‘Relay Support’.
See: ‘Relay Support’.

​
Support

Reach out to us in our Discord server: https://discord.com/invite/fortytwo See the ‘Support’ page for information that can solve your issue.