Give your AI client an MCP reviewer.

The hosted mcplint server exposes one read-only tool: check_mcp_server. It accepts a public HTTPS endpoint or a tools/list snapshot and returns structured scores and findings.

.cursor/mcp.json or ~/.cursor/mcp.json

{
  "mcpServers": {
    "mcplint": {
      "url": "https://mcplint-web.vercel.app/api/mcp"
    }
  }
}

Review Cursor's native confirmation before installing. The button contains only the public endpoint, never credentials.

Install in Cursor

What to ask

Installed-server workflow

Check my installed Grafana MCP server using the MCPLint tool.

If your client exposes Grafana's complete tool definitions to the agent, it should forward them inline as check_mcp_server's snapshot argument — a tools array with MCP name or Cursor-style tool on each entry. File paths and $ref are not supported. Otherwise the agent must ask you for Grafana's public endpoint or a tools/list JSON export. MCPLint never guesses missing schemas.

Review a remote server

“Use mcplint to check https://example.com/mcp. Summarize the three highest-impact fixes.”

Review a snapshot

“Check this tools/list JSON with mcplint and explain why the design score is low.”

Iterate on a schema

“Audit this revised tool surface, compare category scores, and flag overlapping tools.”

Safety boundary

This is static analysis. The hosted tool reads only MCP initialization and tools/list; it never invokes a target tool and makes no LLM calls of its own. Inputs and captured schemas are not saved by the MCP endpoint. For private servers or sensitive credentials, use the local CLI instead.

npx mcp-surface-lint --stdio "node dist/server.js"