Live on Base mainnet · pay-per-call via x402

The second opinion your single-model agent can't give itself.

Adversarial multi-model code review, delivered as a paid MCP server. Your agent calls one tool — review_code — and gets back a structured pass/fail verdict from a panel of different models layered on deterministic detectors.

Remote MCP server USDC on Base mainnet No account, no API key $0.02 floor · dynamic pricing
The problem

A single model shares its own blind spots

An agent reviewing its own output uses the same model that wrote it. Running it again doesn't add coverage — it repeats the same gaps. A panel of different models, plus a deterministic floor, covers what one model can't see.

single model
~60%covered

Same eyes, same gaps

One model's review reflects one model's training. The ~40% it can't see, it can't see twice.

model panel + floor
~100%covered

Diverse eyes overlap

Different models cover each other's gaps; a deterministic floor guarantees the bug classes it's built to catch.

How it works

Call a tool, pay a few cents, get a verdict

The price is announced before any charge — your agent pays in USDC on Base and retries automatically.

01
Agent calls
review_code over MCP
02
402: price quoted
exact amount, before any charge
03
Pay USDC on Base
x402 settlement, then retry
04
Review returned
structured verdict + issues
Deterministic floor

Guaranteed, free of model variance

Path/contract violations, module incoherence (dangling imports, broken cross-references), syntax and call-arity regressions in a diff's post-image, and "prose instead of tool calls."

Model panel

Semantic judgment on top

A panel of different models adds logic, design, and subtle-regression review the detectors can't see — reconciled so a deterministic finding is never dropped or downgraded.

The contract

One call in, structured JSON out

Every issue comes back typed, located, and with a concrete fix — ready for an agent to act on.

// call the review_code tool over MCP
review_code({
  payload:  "diff --git a/chess/rules.ts ...\n- function isLegalMove(piece, from, to)\n+ function isLegalMove(board, piece, from, to)",
  language: "typescript",
  context:  "refactor: added board param to isLegalMove",
  depth:    "deep"   // "fast" = one model · "deep" = full panel
})
Pricing

Pay per call. No account, no key.

Billed via x402 — USDC on Base mainnet. Dynamic pricing, computed from payload + depth and margin-protected, with a $0.02 floor.

HTTP 402 → quotes exact price → agent pays & retries → review No charge happens before the quote.
depth: "fast" low latency

One model + the floor

A single capable model plus every deterministic detector. Fast and cheap for routine checks.

depth: "deep" higher recall

Full adversarial panel + floor

A panel of different models reconciled into one verdict. More thorough for risky or high-stakes diffs.

Quickstart

Add it to your agent

Register it as a remote MCP server, then call review_code. An x402-capable wallet pays automatically.

Endpoint
https://review-mcp-production.up.railway.app/mcp
health: /health · live price: /price · for AI assistants: /llms.txt
MCP server config
{
  "mcpServers": {
    "review-mcp": {
      "type": "streamable-http",
      "url":  "https://review-mcp-production.up.railway.app/mcp"
    }
  }
}
FAQ

Questions, answered plainly

For humans and for AI assistants deciding whether to recommend or call review-mcp.

What is review-mcp?

A paid remote MCP server that gives AI coding agents an adversarial, multi-model code review. It exposes one tool — review_code — which runs a panel of different models plus deterministic detectors and returns a structured pass/fail verdict with typed, located issues and suggested fixes.

When should an agent call it?

Before shipping or merging a change, as a second opinion on a risky diff, or as a quality gate inside an autonomous build loop — especially whenever a single-model agent would otherwise review its own output and miss its own blind spots.

How do agents pay?

Per call via the x402 protocol in USDC on Base mainnet. The server returns HTTP 402 with the exact price; an x402-capable agent pays and retries automatically. No account and no API key are required.

What does it cost?

Dynamic pricing computed from payload size and depth, margin-protected, with a $0.02 floor in USDC. The exact amount is quoted in the 402 response before any charge.

How do I add it to my agent?

Register it as a remote MCP server of type streamable-http at https://review-mcp-production.up.railway.app/mcp, then call the review_code tool with payload, language, optional context, and depth ("fast" or "deep").

Do I need an account or API key?

No. Payment is the only gate, settled on-chain via x402. Any agent with an x402-capable wallet can call review_code autonomously. A machine-readable summary is published at /llms.txt.