smartpolicy

Local

smartpolicy.ioSecurity0Updated 16w ago

About smartpolicy

SmartPolicy provides runtime authorization for AI agents and smart contracts through an on-chain policy registry. It is for teams that need rules separate from enforcement: policies can change through transactions, while its MCP server and REST API issue short-lived signed grants.

Key features

01Register members, conditions, and expiry on-chain
02Check policies against an immutable registry
03Issue short-lived EIP-712 signed grants
04Provide an MCP server and REST API
05Protect contracts with a Solidity modifier

Use cases

01Check whether an agent may act before execution
02Let a backend verify a signed grant
03Change access rules without redeploying contracts

MCP support

SmartPolicy's MCP server offers policy authorization checks to AI systems and can return short-lived signed grants. The server and REST API connect to the on-chain policy registry.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config

{
  "mcpServers": {
    "smartpolicy-protocol-smartpolicy": {
      "command": "npx",
      "args": [
        "-y",
        "@smartpolicy/mcp"
      ],
      "env": {
        "SMARTPOLICY_REGISTRY": "<SMARTPOLICY_REGISTRY>",
        "SMARTPOLICY_VERIFIER": "<SMARTPOLICY_VERIFIER>",
        "SMARTPOLICY_RPC_URL": "<SMARTPOLICY_RPC_URL>",
        "SMARTPOLICY_ISSUER_KEY": "<SMARTPOLICY_ISSUER_KEY>"
      }
    }
  }
}

Run commandnpx -y @smartpolicy/mcp

Environment variables to set

SMARTPOLICY_REGISTRYPolicyRegistry contract address (omit for the public Sepolia default)

SMARTPOLICY_VERIFIERGrantVerifier contract address (omit for the public Sepolia default)

SMARTPOLICY_RPC_URLEVM RPC endpoint (omit for the public Sepolia default)

SMARTPOLICY_ISSUER_KEYSecretOptional grant-issuer private key; enables the grant_issue tool

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages

npm@smartpolicy/mcp0.2.0stdio

Tools

Saved tool list · updated 2w ago

Parameter	Type	Required	Description
action	string	Yes	Action name (e.g. "withdraw") or 0x-prefixed bytes32. Names are keccak256-hashed, matching keccak256("withdraw") in the protected contract.
subject	string	Yes
policyId	any	Yes	Policy ID as decimal string or integer

Details

From the project README

SmartPolicy

Runtime authorization for AI agents and smart contracts.

Define what an agent, a user, or a contract is allowed to do — once, as a policy. Enforce it everywhere: on-chain via a minimal verifier, off-chain via an MCP server and REST API that AI systems consume directly. Update the rules without redeploying anything.

Status (2026-06-12): core contracts live on Ethereum Sepolia (Etherscan-verified, addresses in contracts/deployments/sepolia.json); 54 Foundry tests green; MCP server v0 packaged — npx smartpolicy-mcp runs zero-config against Sepolia, npx smartpolicy-mcp deploy bootstraps any chain from embedded bytecode; AI-consumability proven by unaided fresh-agent execution tests (incl. cold start on an empty chain). NOT audited; internal until mature — no public repo or registry listings yet. Current state, decisions, and how to resume work: PLAN.md.

Why

Every team shipping AI agents hits the same wall: the agent can technically do anything its keys allow, and the only ways to constrain it are hardcoded prompts or redeployed code. Every smart contract team hits the mirror image: access rules baked into the contract at deploy time.

SmartPolicy separates the rule from the enforcement point:

A policy lives in the on-chain Registry: members, admins, conditions, expiry, and mutability flags. It is data, not code. Changing a rule is a transaction, not a redeploy.
Enforcement happens wherever the action happens:
- A smart contract inherits one modifier and checks the Registry (free view call).
- An AI agent (or the service in front of it) asks the MCP server / REST API: "may 0xAgent perform action under policy N?" and receives a short-lived signed grant (EIP-712) that any contract or backend can verify.

What this is NOT

Not a token. There is no protocol token, no ICO, no governance theater. Revenue is fees: a small fee on on-chain policy writes, and metered (x402) pay-per-call on the hosted API. Reads and verification are free.
Not upgradeable. The core Registry is immutable once deployed. Trust comes from code you can read, not admin keys you have to trust. New protocol versions are new deployments.
Not an oracle. SmartPolicy answers "is this allowed?", not "what happened in the world?". Condition values that depend on external facts are attested via signed grants from sources the policy owner chooses.

Architecture (short version)

┌─────────────┐   MCP / REST (x402 metered)   ┌──────────────────┐
│  AI agents   │ ────────────────────────────▶ │ SmartPolicy MCP  │
│  & services  │ ◀──── signed grants ────────  │ server (TS)      │
└─────────────┘                                └────────┬─────────┘
                                                        │ reads + grant issuance
┌─────────────┐   inherit Gate modifier        ┌────────▼─────────┐
│  Protected   │ ────── view calls ──────────▶ │ PolicyRegistry   │
│  contracts   │                               │ (Base, immutable)│
└─────────────┘                                └──────────────────┘

See ARCHITECTURE.md for the full design and the rationale for every decision (chain, fees, trust model, what was deliberately dropped from v1).

Repository layout

contracts/   Solidity core: PolicyRegistry, PolicyGate, interfaces (Foundry)
mcp/         MCP server + REST API (TypeScript, x402 metering)
sdk/         TypeScript client SDK (@smartpolicy/sdk) — reads, grants, tx-building
docs/        Integration guides (planned)

Lineage

SmartPolicy is a clean rewrite of the strongest ideas from a 2025 prototype ("Smart Policy Protocol", Sepolia): the policy registry, the tiered integration model, and hybrid on-chain + signed-token enforcement. The rewrite deliberately drops the SPOL token/ICO/governance suite, the sports-prediction oracle, the factory contracts, and the upgradeable-proxy trust model, and fixes the known security issues (tx.origin checks, missing reentrancy guards, header-only API auth).

License

Apache-2.0 — see LICENSE.

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

Read the getting-started guide

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.