AI Proof of Us is an MCP-first receipt protocol for humans working with AI.
It starts from a simple belief: when people spend real hours building, debugging, researching, writing, and coordinating through AI agents, that work should be able to leave a private, portable receipt.
Private, signed work receipts for AI agents. Local-first MCP server with EIP-712/Ed25519 proof, replay protection, and optional AIPOU rewards on Base.
AIPOU gives Codex, Claude, Cursor, OpenClaw, local models, and other MCP-compatible clients a shared way to create signed, privacy-preserving receipts for AI-assisted tasks.
The human reward loop is the point: people can keep receipts for the work they do with AI all day, and validator-approved receipts can claim AIPOU on Base. The protocol exists to make that claimable work more honest, portable, and useful across agents.
The developer surface is intentionally small: receipts, hashes, lifecycle hooks, and workReceiptId interoperability. Agents, marketplaces, and services can also reference AIPOU receipts or voluntarily accept AIPOU as settlement when both sides agree to use it.
The first version ships as:
an MCP server that records privacy-preserving AI task receipts
an installable OpenClaw skill for agent workflows
docs for MCP clients, agent builders, lifecycle adapters, anti-abuse, and future attestation design
a reward model that converts valid receipts into claimable emissions
an ERC-20 reward token on Base: AI Proof of Use (AIPOU)
The core idea is simple:
agent starts task -> MCP creates nonce -> AI work happens -> signed receipt -> validator checks -> optional AIPOU claim
Start here if you are building or testing an agent integration:
AIPOU and AI Proof of Us are currently unregistered project identifiers. The MIT license covers the software, not permission to present another token, service, or package as the official AIPOU project.
This is not "AI usage mining" and it is not meant to reward raw prompt spam or attract passive token speculation. AIPOU is for humans doing real work with AI and for developers who want to test whether that work can produce portable receipts across agent clients.
In plain language:
Humans work with AI -> agents create receipts -> approved receipts claim AIPOU -> agents and marketplaces may accept AIPOU as settlement.
Base mainnet deployment
Contract: 0x55f0Cc5e51A1284D20337d6cbb18938C8A1ABCbB
Chain: Base Mainnet (8453)
Token: AI Proof of Use (AIPOU)
Supply: 100,000,000 AIPOU
Cap: 1,000,000,000 AIPOU
Logo: assets/token/aipou.png
The pool was initialized with minimal ETH liquidity and later deepened to
approximately 99.23M AIPOU + 0.02957 WETH. A total of
1,712.89145801936861611 LP tokens, representing 99.9999% of the LP supply,
are held by an immutable lock until July 8, 2027. The lock has no owner, admin,
or early-withdrawal path; after expiry, anyone may trigger release, but the LP
tokens can only return to the fixed beneficiary.
Liquidity is still small and its price remains volatile. The lock prevents the
protocol wallet from removing the locked position during the published period;
it does not guarantee token value, trading volume, or future liquidity.
The initial experimental pool buy was later returned through Aerodrome so the large early purchase would not remain as a misleading concentrated holder balance. See AIPOU Tokenomics and Launch Transparency.
Repository layout
contracts/ ERC-20 token, deploy scripts, Hardhat tests
mcp-server/ MCP server that records AI task receipts
docs/ architecture, anti-abuse model, Base launch notes
skills/ installable agent skills, including OpenClaw
huggingface-space/ static public protocol explainer
Token
Name: AI Proof of Use
Symbol: AIPOU
Chain: Base
Cap: 1,000,000,000 AIPOU
The token emission controller is AIPOUClaims. It mints only receipts included in a validator-published Merkle root and rejects a receiptId after its first claim.
AIPOU claims are optional settlement for approved receipts. They do not prove hidden AI use, objective task value, provider endorsement, provider inference without cryptographic provider evidence, or security-policy compliance.
MCP server
The MCP server exposes tools for creating task receipts:
get_aipou_contract
get_aipou_identity
estimate_ai_reward
begin_ai_task
complete_ai_task
export_ai_receipts
get_aipou_status (recorded, pending, claimed, and onchain balance summary)
settle_ai_rewards (one limited validator batch)
settle_all_ai_rewards (one-command claim path for all pending eligible receipts)
Receipts store hashes and metadata, not raw prompts or model outputs.
The dedicated farming key signs EIP-712 authorizations locally. Never paste it into a chat or use a primary wallet. The collector has a separate Ed25519 key that cannot move funds.
Frameworks do not need to understand Merkle trees, Base, or token claims to integrate the receipt layer. The minimal adapter watches task start and task end, records provider/model metadata and hashes, and exposes the AIPOU receiptId as workReceiptId for workflow metadata, traces, UI, audit exports, payment/session records, or later optional settlement.
Quick start
Create A Verified Local Receipt
This is the lowest-friction adoption test — one command, no checkout, no setup:
npx -y aipou-mcp-server --demo
It creates an ephemeral wallet, records and cryptographically verifies one local
receipt, prints the workReceiptId integration object, and removes all temporary
state. It requires no wallet setup, funds, network access, claim, or raw prompt
and output storage.
Create A Persistent Dedicated Identity
After evaluating the disposable demo, initialize a dedicated farming wallet
without printing its private key:
npx -y aipou-mcp-server --init
The command creates ~/.aipou/agent-wallet.key with restricted file access and
prints a ready-to-copy MCP configuration that references the key by path.
Re-running it reports the existing identity instead of overwriting the key.
Verify the persistent setup without writing files, contacting an RPC, moving
funds, or submitting a claim:
npx -y aipou-mcp-server --doctor
The diagnostic returns structured checks for Node.js, storage, identity,
contract addresses, and chain. It prints only the public wallet address, never
the private key.
From a source checkout, the same check runs with:
npm install
npm run demo -w mcp-server
Test The Receipt Adapter
This is the fastest path for maintainers and agent-framework builders. It creates a local receipt with an ephemeral wallet and prints the workReceiptId object that a framework can attach to run metadata, traces, audit exports, or payment/session metadata.
npm install
npm run build -w mcp-server
cd examples/lifecycle-adapter
npm install
npm run demo
No claim is made. No funds move. No raw prompt or output is uploaded.
Run The MCP Server
Install dependencies:
npm install
Build all packages:
npm run build
Run contract tests:
npm run test -w contracts
Run the MCP server locally:
npm run dev -w mcp-server
MCP clients can launch the published server with:
npx -y aipou-mcp-server
Release [email protected] is live on npm and marked latest in the official MCP Registry. It was published from GitHub Actions through npm Trusted Publishing with SLSA provenance. See docs/npm-publication.md.
Local MCP config example
Keep the farming private key in a .env file next to the repo and point the server at it with DOTENV_CONFIG_PATH. Never paste the private key into the MCP client config itself: client configs are often synced, logged, or read by other tools.
Prefer AIPOU_AGENT_KEY_FILE, generated by --init, over embedding
AIPOU_AGENT_PRIVATE_KEY in a client configuration. The .env file can still
hold the private key for backward compatibility, plus AIPOU_CLAIMS_ADDRESS.
On the validator machine, keep the validator key out of that shared .env:
store it in a separate file and reference it with AIPOU_VALIDATOR_KEY_FILE so
farming-only processes never load it.
Reward flow
The agent calls begin_ai_task and signs a unique nonce with its farming wallet.
The client collects usage and calls complete_ai_task with the output hash.
The MCP derives the trust tier and signs the receipt with Ed25519.
The validator rejects repeated nonces and repeated task/output evidence.
get_aipou_status lets the user see how much has already been recorded, claimed, and left pending.
After an explicit settlement request, settle_all_ai_rewards processes all currently eligible pending receipts from the shared AIPOU_DATA_DIR in bounded batches.
AIPOUClaims rejects claimed receipt IDs and mints AIPOU to each farming wallet.
Users should normally ask for status first, for example show my AIPOU status, to see already claimed rewards, pending receipts, and the farming wallet's onchain balance. An explicit request such as claim my AIPOU is the trigger for settlement only when pending eligible receipts exist. Broad claim requests use settle_all_ai_rewards; settle_ai_rewards remains available for a single limited batch. The MCP client and its user keep the final say on how on-chain transactions are confirmed; the server never asks a client to skip its own confirmation policy. The validator can optionally enforce a settlement policy (minimum work floor per receipt via AIPOU_MIN_RECEIPT_TOKENS, per-wallet daily receipt limit via AIPOU_MAX_DAILY_RECEIPTS_PER_WALLET); both checks are disabled by default.
The trust tier is derived by the MCP and recomputed by the validator. Users cannot self-report provider_signed; a provider tier requires a valid provider signature from a configured public key.
No tool list yet. Remote servers get one after the first health check.
FAQ
Questions
How do I install AI Proof Of Us?
Copy the stdio config for Cursor, Claude Desktop, Claude Code, or VS Code on this page. It uses the publisher's install, not a PlugMCP proxy.
Does AI Proof Of Us need an API key?
The publisher lists these environment variables: AIPOU_AGENT_PRIVATE_KEY, AIPOU_AGENT_KEY_FILE, AIPOU_CONTRACT_ADDRESS, AIPOU_CLAIMS_ADDRESS, AIPOU_RPC_URL, AIPOU_DATA_DIR.
What transport does AI Proof Of Us use?
AI Proof Of Us runs locally over stdio.
When was AI Proof Of Us last checked?
Last checked 2026-10-26. Private, signed work receipts for AI agents. Local-first MCP server with EIP-712/Ed25519 proof, replay protection, and optional AIPOU rewards on Base.