Prismnetwork Tech

io.github.prismnetwork-techPublished by GitHub account prismnetwork-techv0.10.0Updated Oct 25, 2026

Rent real NVIDIA GPUs from your agent. Browse with no wallet; pay per second onchain.

The publisher's own server: Tech.eu Funding Explorer

VerifiedSTDIODesktop onlyCloud & InfrastructureFinanceAI & ML

Overview

AI-generated overview

Lets an assistant browse and lease real NVIDIA GPUs through Prism Network, paying per second onchain, and run commands or paid inference on them.

What you can do
Browse live GPU capacity and pricing, and check the price index and public receipts without a wallet. Check the agent wallet's address and balances, the enforced spending limits, and the wallet's leases. Lease a GPU and keep it with SSH access, or lease and run a one-shot command; run commands on an existing lease and release it to stop billing. Buy single or batch LLM generations, or confidential generations inside a GPU TEE, and verify their attestations. Seal private items in a wallet-derived vault, list and read them, release them into a lease, or delete them.
When to use it
Use it when an assistant needs real GPU compute for training, benchmarking or batch jobs, or paid LLM inference, without setting up cloud accounts by hand. Browsing capacity and prices is free, so it is also useful just to compare GPU costs.
Requirements
Runs locally via npx as a stdio process, so Node.js is needed. Leasing and inference need PRISM_AGENT_KEY, a wallet private key, plus PRISM_ESCROW; the wallet must hold USDG and Robinhood-Chain ETH for gas. Optional: PRISM_DAILY_BUDGET_USDG, PRISM_MAX_USDG, PRISM_SPEND_POLICY. Provisioning can block for one to four minutes, so long tool calls must be allowed.
Before you install
The server holds a wallet private key in PRISM_AGENT_KEY and spends real funds onchain; fund a dedicated wallet with only what you can afford to lose. Default ceilings are 1 USDG per lease or generation and 5 USDG per rolling 24 hours, and setting PRISM_DAILY_BUDGET_USDG to 0 removes the daily ceiling. Spending tools are marked destructive and ask for confirmation. Vault items are sealed under a wallet-derived key, and releasing one into a lease exposes it to that host.

Installation

In SourceWeft

  1. Open Prismnetwork Tech in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

@prismnetwork/mcp

An MCP server that lets Claude (or any MCP client) see and lease real GPUs through Prism Network. Give it a wallet and it handles auth, onchain payment, provisioning, and SSH.

Try it without a wallet

Looking costs nothing and needs no configuration:

sh
claude mcp add prism -- npx -y @prismnetwork/mcp

Then ask what you can rent. prism_list_gpus answers from live capacity with real prices. Leasing spends money, so those tools ask for a wallet and say so.

Tools

ToolWallet
prism_budgetno
prism_list_gpusno
prism_price_indexno
prism_receiptsno
prism_walletyes
prism_leasesyes
prism_inferyes
prism_infer_batchyes
prism_confidential_inferyes
prism_verify_attestationno
prism_lease_and_runyes
prism_leaseyes
prism_runyes
prism_batch_runyes
prism_end_leaseyes
prism_vault_storeyes
prism_vault_listyes
prism_vault_readyes
prism_vault_deleteyes
prism_vault_releaseyes
  • prism_budget: the spending limits this server enforces, what it has spent in the last 24 hours, and the recent charges.
  • prism_wallet: the agent's address and USDG/ETH balances.
  • prism_list_gpus: GPUs available to lease, with price per second and per hour.
  • prism_price_index: sourced and settled pricing per GPU model, for cost estimates.
  • prism_receipts: recent settled receipts from the public proof feed, with the settlement transaction on Robinhood Chain.
  • prism_leases: this wallet's leases and their state.
  • prism_infer: buy one LLM generation from the managed inference endpoint, paying the quoted USDG price from this wallet (about 0.01 USDG). Waits through a cold start; an unconsumed payment is kept and reused on the next call instead of paying twice.
  • prism_infer_batch: buy many generations in one paid call, at the single-generation price times the number of prompts. Each prompt runs whole on a rented GPU and they are spread across every GPU the endpoint holds, so a list of prompts finishes far sooner than the same prompts sent one at a time. Returns every answer in order with a Merkle receipt naming the leases that did the work.
  • prism_confidential_infer: buy one generation that runs inside a GPU TEE. The message contents are encrypted to a key the enclave's own attestation commits to, so this server's relay carries ciphertext and cannot read the prompt or the answer. Returns the answer, the cost and a receipt id the workload signed over the exact bytes of the exchange.
  • prism_verify_attestation: check one of those generations against its receipt and the hardware behind it. The server remembers the digests of the last few confidential calls, so the verdict is bound to the bytes it actually sent and received. Every check is reported with its result, including the two that cannot be established today: private-key custody, and where TLS terminates.
  • prism_lease_and_run: lease a GPU, run a command, return the output (one shot).
  • prism_lease: lease a GPU and keep it; returns a lease_id and SSH access. The SSH block carries host_key_fingerprint and host_key_claim when the network can say which machine should answer, and says so plainly when it cannot. Check it before connecting by hand; prism_run checks it for you.
  • prism_run: run a command on an existing lease.
  • prism_batch_run: fund a lease that runs one command with no interactive access; the node reports the signed output. Matches only suppliers at trust class isolated or above, so it can find no supplier when none is online.
  • prism_end_lease: release a lease. Billing stops at the release; the unused deposit returns after settlement.
  • prism_vault_store: seal private data under the wallet-derived key.
  • prism_vault_list: list sealed items; values are never returned.
  • prism_vault_read: decrypt one item in this process.
  • prism_vault_delete: permanently delete an item.
  • prism_vault_release: authorize an item into a lease that clears its trust floor.

Spending limits

Two ceilings bound what this server can spend, and a refusal quotes both.

VariableDefaultWhat it bounds
PRISM_MAX_USDG1Any single lease or generation. max_usdg on a call may lower it, never raise it past this.
PRISM_DAILY_BUDGET_USDG5Everything in a rolling 24 hours. 0 removes the ceiling.
PRISM_LEDGER_PATH~/.prism/spend.jsonWhere the spend is recorded.

A max_usdg above PRISM_MAX_USDG is clamped back to it. The argument is written by the agent being bounded, so it is treated as a request for a lower ceiling and never as permission for a higher one.

Spend is written before the money moves, so a crash between funding an escrow and answering is counted rather than forgiven, and a restart does not hand back a fresh day's allowance. Only an attempt that provably never reached the chain is reverted. Two clients pointed at one wallet share one ceiling, because they share the ledger file.

None of this is the real limit. Fund a dedicated wallet with what you are willing to lose: that balance is what survives a bug in everything above.

Spend policy

The limits above cap how much an agent spends. A spend policy decides whether a given lease has a reason you accept. Set PRISM_SPEND_POLICY to the policy as JSON, or to the path of a JSON file:

json
{  "policy_id": "gpu-v1",  "allow": ["fine_tune", "benchmark"],  "require": ["needs_gpu"],  "minimums": { "needs_gpu": 0.8 }}

Every lease tool then takes a decision: the action, what made the call, and typed answers with a confidence from 0 to 1. A decision that misses the policy is refused with every reason before anything is quoted, funded or counted against the budget. One that passes is hashed into the escrow deposit, so the lease on chain carries a reference only that decision reproduces. prism_budget shows the policy in force. Without PRISM_SPEND_POLICY a decision is optional and still binds when given.

The binding shows a decision existed before the money moved and met your rules. It does not show the decision was sound.

Tools that spend are annotated destructiveHint and carry anthropic/requiresUserInteraction, so Claude Code asks before every one of them even in modes that otherwise approve tools automatically.

Vault

An agent that handles a card, an identity document or a credential should not write it into a leased workspace. prism_vault_store seals it under a key derived from the agent's wallet inside this server process; Prism receives ciphertext and holds no way to read it.

Each item names the weakest workspace trust class it may ever be released into, and new items default to confidential, above anything the network serves today. prism_vault_release is therefore refused on current capacity instead of handing a secret to a host that can read it. Lowering an item's floor is a deliberate act, and allowed releases are recorded against the account.

Configure

Point your MCP client (Claude Desktop / Code) at the published server:

json
{  "mcpServers": {    "prism": {      "command": "npx",      "args": ["-y", "@prismnetwork/mcp"],      "env": {        "PRISM_AGENT_KEY": "0x<agent wallet private key>",        "PRISM_ESCROW": "0xfD4228eEEfC49e4b76A0CD40af9fdd546220B2FD"      }    }  }}

The wallet needs USDG (0x5fc5360D0400a0Fd4f2af552ADD042D716F1d168) and Robinhood-Chain ETH for gas. See the SDK's Funding section for how to fund a fresh wallet.

Or add it to Claude Code in one line:

sh
claude mcp add prism \  --env PRISM_ESCROW=0xfD4228eEEfC49e4b76A0CD40af9fdd546220B2FD \  --env PRISM_AGENT_KEY=0x<agent wallet private key> \  -- npx -y @prismnetwork/mcp

Timing

prism_lease and prism_lease_and_run block while a GPU provisions (usually one to four minutes, occasionally longer on a slow host). Configure your MCP client to allow long tool calls.

Source: mcp/README.md at commit 4e32eb5

Tools

0
SourceWeft has not started this package in its sandbox yet, so its tools are not known.

Version history

2
  1. v0.10.0LatestSep 27, 2026
  2. v0.9.4Sep 16, 2026

Score

50% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts

Other Tech integrations