An agent can run a firm here.
DecentralProp speaks MCP. An agent can read every firm's solvency, verify that a payout actually landed, and deploy a funding firm of its own that real traders can buy from. The reading half needs no credentials at all.
Two halves, and only one of them costs anything
The server splits cleanly. Twelve read tools answer questions about the protocol and need no wallet, no key and no account. Five write tools deploy and disclose a firm, and those need a funded wallet.
Read: free, safe, and useful on its own
Every read tool calls a route that was already public. Nothing is being opened up for agents; the transparency surface has always been readable by anyone. A protocol that claims settlement is verifiable only means it if something verifies it, and agents are more patient about that than people are.
Write: a real firm, with real money
Deploying charges a one-time launch fee in SOL and cannot be reversed. The moment it lands, the storefront is live and can sell evaluations to real people. Three separate catches stand between an agent and that transaction, and they are described below.
The twelve read tools
The one worth knowing about is get_firm_solvency. A prop firm can look excellent on pass rate and payout history and still be unable to pay you. Solvency reads the three-tier waterfall and says what is actually realizable against what is actually owed.
list_firmsEvery firm in the directory, with its trust score, pass rate and delivered payouts. Most other tools take a firm from this list.
get_firm_solvencyThe three-tier payout waterfall: what is realizable against what is owed. The question that actually matters before buying an evaluation.
get_firm_payoutsPayouts a firm has delivered on-chain, each with its transaction signature. Settled facts, not claims.
verify_payoutThe settlement certificate for one payout. Use it to check a claim instead of believing it.
get_settlement_proofThe committed roots and fraud-proof window behind a single evaluation, so its result is checkable rather than asserted.
get_firmOne storefront in full: branding, terms, evaluation plans, and whether an agent operates it.
get_firm_transparencyA firm's on-chain state as the indexer mirrors it: treasury, tier, exposure, configuration.
get_trader_recordOne wallet's record across the protocol. A trader is a wallet, so two firms do not split one person in half.
get_trader_challengesEvery evaluation a wallet has run, with phase, status and outcome.
get_protocol_payoutsThe protocol-wide payout wall, newest first. Anonymised but present.
get_leaderboardTraders ranked by payouts that landed on-chain. Opt-in, so it ranks the top earner who published a profile.
get_chain_integrityThe watchtower's view of settlement-chain integrity: whether any committed batch root is disputed or missing.
Connecting an agent
Any MCP client. Point it at the server, and the read tools are available immediately.
{
"mcpServers": {
"decentralprop": {
"command": "npx",
"args": ["-y", "@decentralprop/agent-mcp"],
"env": {
"DECENTRALPROP_BASE_URL": "https://api.decentralprop.com"
}
}
}
}The package is not on the public registry yet, so that command will not resolve today. The server ships with the protocol and the configuration above is its real shape. If you are building an agent against DecentralProp now, get in touch and we will get you the build.
| Variable | Effect |
|---|---|
DECENTRALPROP_BASE_URL | Gateway origin. Defaults to production. |
DECENTRALPROP_RPC_URL | Solana RPC. Only the write tools use it. |
DECENTRALPROP_WALLET | Path to a keypair file. Supplying it is what enables the write tools at all. |
DECENTRALPROP_ALLOW_SPEND | Set to 1 to let deploy_firm actually spend. Without it, only previews run. |
The wallet is read from a file path, never a raw key in an environment variable. Environment variables turn up in process listings, shell history and crash reports. A file with restrictive permissions does not.
What deploying actually does
Seven steps. Six of them are free and reversible. One is not.
1. Sign in
The agent signs a server-issued nonce with its wallet key. Ordinary ed25519, no browser. That signature is the whole of the identity, and the wallet it proves becomes the firm's owner.
2. Declare the operator
The wallet registers as agent-operated before the firm exists, so the storefront carries the disclosure from its first render rather than from a backfill nobody triggers.
3. Validate the draft
Name and storefront address are checked against every other firm. A collision fails here, where failing is free, instead of after the money is gone.
4. Stage it
The draft is stored durably and has to confirm before anything is spent. A firm that deployed on-chain with its name lost is the failure this step exists to prevent.
5. The keeper co-signs
The protocol builds one transaction carrying the firm, its treasury, its vaults and the launch fee, signs its own half, and hands it back.
6. The agent signs
Balance is checked first, so an underfunded wallet gets a number instead of a failed simulation. Then the second signature goes on, and the firm exists.
7. The storefront opens
Name, address and evaluation plans provision from the staged draft. The firm can sell.
The five write tools, and what stops them
Run preview_deploy_firm first. It does everything except sign, and the launch fee it reports is the gateway's own quote rather than an estimate.
check_firm_specValidate a firm spec offline. Catches shape problems and out-of-band evaluation prices without contacting anything.
preview_deploy_firmA full dry run against the live gateway: real sign-in, real name and address validation, a real keeper-built transaction, a real balance check. Then it stops without signing.
deploy_firmThe real thing. Signs a transaction that spends the launch fee, and cannot be undone.
register_agent_operatorDeclare that an agent runs this wallet's firm, so the storefront says so publicly.
get_deploy_tiersWhat each deploy tier costs.
Three catches, each independent
- No wallet configured and the write tools do not register at all. An agent pointed at the server without one cannot see them, let alone call them.
- No explicit spend flag and
deploy_firmrefuses before it touches the network. - An explicit acknowledgement is required by the tool's own input schema, so a deploy can never happen as a near miss of a preview.
Autonomous up to the money
Signing a login nonce and signing a transaction are not the same act, so the SDK keeps them separate. One proves who the agent is and authorises nothing. The other spends the launch fee.
The agent holds the key
Simplest, and correct when the wallet is funded for the agent specifically and you accept that it can spend the balance. Nothing blocks, nothing waits.
A human approves the money
The agent signs its own logins and runs the whole flow unattended, then hands the transaction to whatever you decide should approve it: a person, a multisig, a policy engine. It never signs that one itself.
This costs the protocol nothing to support. The deploy transaction is already built and co-signed server-side and handed back for a second signature, so the only question is who provides it.
Why an agent can operate a firm here
Because an agent operator cannot do anything a human operator could not. On a centralized prop firm the operator writes the rules, grades you against them and profits when you fail. Here the protocol holds all three, and it does not care who deployed the firm.
The evaluation rulebook
Profit target, daily loss, total drawdown and minimum trading days are platform-set and server-enforced. An operator cannot make its own evaluation harder or its rules ambiguous.
Leverage
Set by the platform and reduced by the risk engine under stress. A draft that asks for its own leverage is ignored.
Payouts
The waterfall runs on-chain. A funded trader is paid by the protocol's rules, not by an operator's willingness.
The fee ladder
The affiliate fee, the backstop premium and every fee leg are fixed protocol-wide and enforced by the programs, not by the firm.
The risk engine
The ARE watches every firm continuously and moves its terms within a band. It answers to the protocol, and it treats an agent-run firm exactly like any other.
The one thing an operator chooses
A profitability stance. It sets the funded profit split and a small evaluation-price multiplier, and the risk engine moves the firm within a band from there. That is the entire economic surface, for an agent and for a person alike.
Max Profit
The trader keeps 75% of funded profit. Slightly better for the firm, and a modestly higher evaluation price.
Balanced
The trader keeps 80%. The simulation-validated middle, and the default.
Conservative
The trader keeps 85%. Easier for traders and better for attracting volume.
Every agent-run firm says so
A firm deployed through these tools is labelled agent-operated, on its storefront and in the firm directory. Traders send real SOL to these storefronts and are entitled to know.
The label is narrow and true. An agent picks the profit split and writes the storefront copy. Evaluation rules, leverage and payout enforcement come from the protocol and are identical at every firm on it.
There is no counterpart badge for human operators, deliberately. Nothing verifies that, and a label claiming it would be the only dishonest thing on the page. A firm without the agent label has not declared one, which is not the same as a verified person.
What it costs
One on-chain fee, charged in SOL at the live rate. No monthly rent, at any tier.
| Tier | One-time launch fee |
|---|---|
| Starter | $1,000 |
| Growth | $5,000 |
| Pro | $10,000 |
| Scale | $25,000 |
| Enterprise | $50,000 |
What an agent cannot do
Trade. Trading on DecentralProp is UI-only by design and has no programmatic surface, for agents or for anyone else. An agent can own and operate a firm. It cannot place an order, on its own firm or on any other.
