Rein

What the Rein MCP server does

The Rein MCP server brings Rein’s demand-side payment guard to MCP clients. An agent can use a governed fetch for resources protected by x402 paywalls, while the guard evaluates the intended payment against configured spending policy before funds move. The server also exposes read-only visibility into the rules and authority governing that agent.

Rein is non-custodial. It does not hold the agent’s funds or give the MCP client a way to expand its own authority. The design prevents the connected agent from approving its own escalation, changing policy, disabling a freeze, or creating a signing key through the tool surface.

Use this server when an agent needs to call paid HTTP resources through x402 and the surrounding application should enforce spending limits and retain payment observability. It is intended for MCP-capable agent harnesses and avoids requiring changes to the agent’s existing fetch implementation.

How it works

The MCP client connects to the Rein guard and acts as the agent on the other side of the protocol pipe. When the agent requests a paid resource, Rein evaluates the payment request under its policy system before allowing the transaction to proceed. Decisions are signed with Ed25519 and added to a SHA-256 hash-chained audit log by the underlying policy engine.

The guard can run against fully offline mock payment rails or against real x402 rails on Base Sepolia. In the live configuration described by the project, payments use EIP-3009 USDC authorization and the hosted x402.org facilitator. The mock option is useful for testing without accounts, Docker, or a blockchain.

The Rein MCP server provides the agent with governed payment access and read-only inspection. It does not expose administrative controls that would let the agent alter its own restrictions. Rein’s broader stack can also reconcile payments and identify spending that bypassed the guard, but those capabilities belong to the surrounding Rein packages rather than being described as additional MCP controls here.

Setup and configuration

Install and run the published MCP package with:

npx -y @reinconsole/mcp

Connect that command to an MCP-capable harness such as Claude Code or a Codex-class agent. The README presents the MCP package as a way to add governed fetch behavior without application code changes. Rein MCP server deployments can use the project’s offline mock rails for local work, or the Base Sepolia x402 path for live testing.

The repository states that the packages require Node.js 22 or newer and are released under the @reinconsole npm scope. The supplied material does not document MCP-specific environment variables, client configuration fields, or a separate account setup process, so those details should be confirmed in the package documentation before deployment.

Tools and capabilities

The documented MCP-facing capabilities are:

• Perform a fetch through Rein’s spend-governed x402 path.
• Check a paywall against the agent’s payment policy before payment.
• Inspect the rules and authority applied to the connected agent in read-only form.
• Produce observable, signed payment decisions through the underlying guard.
• Keep the agent from editing policy, approving escalation, unfreezing itself, or minting a key through MCP.

Limitations and notes

Rein MCP server access governs payment authority; it is not a wallet or custody service. The README describes live x402 operation on Base Sepolia, not unrestricted production-chain support. The MCP server’s documented inspection surface is read-only, and the supplied material does not list individual MCP tool names or detailed request and response schemas.

The project identifies the MCP package as part of a v0.3 stack, while the npm installation section says the latest published release is 0.5.0. Check the changelog when upgrading across the documented 0.2.0-to-0.3.0 boundary. The project reports Node.js 22 or newer as a requirement.