From discovery to a running tool

Search the registry from the browse UI, or query the public API. Results are quality- and safety-scored, so you can filter for servers worth running.

# Find GitHub-related servers scoring 70+ on quality
curl "https://mcprating.io/api/v1/servers?search=github&minQuality=70"

The same data is available as an official-format subregistry at /v0.1/servers.

The Gateway is itself an MCP server. Point your MCP client (Claude Desktop, Cursor, or your own agent) at it once, and it becomes a single endpoint to discover, connect, and proxy every other server — each run in a sandbox.

{
  "mcpServers": {
    "mcprating": {
      "command": "npx",
      "args": ["-y", "@mcprating/gateway"]
    }
  }
}

From an agent, the Gateway exposes meta-tools. Discover from the registry, connect a server (spawned in a sandbox), then call its tools by name — every call proxied and recorded in the audit trail.

import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StdioClientTransport } from "@modelcontextprotocol/sdk/client/stdio.js";

const transport = new StdioClientTransport({
  command: "npx",
  args: ["-y", "@mcprating/gateway"],
});
const agent = new Client({ name: "my-agent", version: "1.0.0" });
await agent.connect(transport);

// 1. Discover from the MCP Rating registry
await agent.callTool({ name: "mcp_discover", arguments: { query: "github", limit: 5 } });

// 2. Connect a server — spawned in a sandbox
await agent.callTool({ name: "mcp_connect", arguments: { slug: "github-mcp-server", confirmed: true } });

// 3. Call its tools — proxied, sandboxed, audited
await agent.callTool({
  name: "github-mcp-server__search_repositories",
  arguments: { query: "model context protocol" },
});