mcpField
M
MIZAN · Proof before payment

MIZAN · Proof before payment

Remote
mizan.marketSecurity0Updated 6d ago

About MIZAN · Proof before payment

MIZAN is a remote verification service for agents and organizations, aimed at parties assessing a counterparty before payment or other commitments. It returns signed allow-or-refuse decisions and lets agents prove records without disclosing their methods.

Key features

  1. 01Signed ALLOW or REFUSE checks
  2. 02Checks against custom requirements
  3. 03Private performance-record proofs
  4. 04Agent discovery by capability
  5. 05Payment checks with x402 guard

Use cases

  1. 01Assess an agent before paying or hiring
  2. 02Require a published standard before allocation
  3. 03Prove performance without exposing strategy

MCP support

MIZAN exposes verification, proof and discovery tools through a remote MCP server, with a local stdio option. Clients can check identities and requirements or request proofs.

Configuration

Remote servers need no local install. Pick your client, then copy the config or install in one click.

Endpointhttps://studio.mizan.market/mcp
Copy install command
Claude CodeTerminal
Cursor.cursor/mcp.json
VS Code.vscode/mcp.json
Claude Desktopclaude_desktop_config.json
Windsurf~/.codeium/windsurf/mcp_config.json

Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.

Tools

6
ParameterTypeRequiredDescription
sha256stringNo
credential_idstringNo
reverify_locallybooleanNo

Details

From the project README

MIZAN MCP: check before you pay. Prove to get paid.

mcp-name: io.github.muaviamohammed/mizan

No capital moves to a machine without proof. Before your agent pays, hires, copies or allocates to another agent, one call returns a signed ALLOW or REFUSE on what that agent claims, against your own bar if you set one. And any agent can prove its own record without revealing its model, code, prompts or strategy.

claude mcp add --transport http mizan https://studio.mizan.market/mcp

Remote, stateless, no install, no key needed to check. Listed on the official MCP registry.

Check. Prove. Require. Find.

ToolWhen to call it
check(anything)Before money moves. Pay, hire, copy or allocate: one signed ALLOW or REFUSE, free, in milliseconds. anything = an agent id, a company domain, an API, MCP or A2A endpoint, a GitHub, Hugging Face, npm or PyPI identity, a public key or a wallet. A REFUSE carries a ready message the other side uses to prove itself. Add requirement="req_…@1" to check against a published bar, clause by clause.
prove()To get paid. What your record really earned, how accurate your forecasts are, how reliable your service is. Free private preview first; your model, code, prompts and strategy never leave your machine.
more()Everything else: require(text) (write your bar in plain words), find(capability, bar) (only proven agents that meet it), link() (one proof across every identity you hold), more('accountable') (who an agent acts for, what happened after others paid it, and alerts when its status changes), receipts, the open standard, the public log.

Your first tool call registers you as a MIZAN agent: a private record and a key. Nothing about you is public unless you choose. The default address lists three tools; every tool stays callable there, and the full list is at https://studio.mizan.market/mcp/full.

Why an agent can rely on the answer

  • Proven, not reported. The record is re-run inside a STARK proof under the open, frozen VTR-1 standard: after costs, on committed data, out-of-sample, deflated for every try.
  • It says no. Refusals are published beside passes, MIZAN's own AI models among them. That is what makes a pass worth something.
  • Don't take our word for it. Every reply is Ed25519-signed (key at https://studio.mizan.market/.well-known/mizan-keys.json). Every verdict is a leaf of a Bitcoin-anchored transparency log. The verifier re-runs any credential offline on macOS, Linux or Windows: exit 0 or it is refused.
  • Your method stays yours. Never send a strategy, model, weights, code or prompt anywhere. Positions only.

In the payment itself (x402)

import { guardFetch } from "https://mizan.market/skill/mizan-x402-guard.mjs";
const pay = guardFetch(fetchWithPayment, { mode: "advise" });   // later: { mode: "enforce" }

The guard reads the seller's 402 terms, checks the wallet being paid, then pays. Advise mode attaches the signed answer; enforce mode pays only on ALLOW. Python: from mizan.x402guard import guard. Platforms: the require kit adds the Proven / Refused / Unproven label next to every agent in two lines of HTML.

Local (stdio)

pip install "mcp>=2"
python3 mizan_mcp_server.py
# Claude Code: claude mcp add mizan -- python3 /path/to/mizan_mcp_server.py

Also available as

  • A2A agent: https://studio.mizan.market/a2a; card at https://mizan.market/.well-known/agent-card.json.
  • Framework tools: LangChain, LangGraph, CrewAI (mizan-agent-tools), elizaOS (plugin-mizan), TypeScript (mizan-agent-tools-js).
  • CLI: mizan verify <record-or-wallet> previews any Hyperliquid, Freqtrade, MT5, QuantConnect or TradingView record in one step (wheel).

Links

mizan.market · Check any agent · Registry · Skill for agents · [email protected]

Every API key is a referral key: agents that bring agents earn, credited only on a real mint (GET /api/rewards/program).

A credential is a seal on a past record, not a forecast. MIZAN never holds or routes funds.

MIT. © 2026 Mizan Verification, Inc.

Keywords: AI agent verification · check before you pay · proof of performance · verifiable track record · zero-knowledge proof · STARK · RISC Zero · MCP · A2A · x402 · ERC-8004 · backtest overfitting · deflated Sharpe ratio · trading bot due diligence

Comments

Markdown supported

New to MCP?

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

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.