Model Context Protocol

MCP Server & Tools

Expose your indxBASE project as an MCP server. Any compliant AI agent — ChatGPT, Claude, Cursor — can query your data securely.

Connect an AI agent

Connect

ChatGPT

Add as a custom GPT action.

Connect

Claude

Paste the MCP URL into Claude Desktop.

Connect

Cursor

One-click add to .cursor/mcp.json.

Custom agent

Any client speaking MCP over HTTP/SSE.

Available MCP tools

.cursor/mcp.json

{
  "mcpServers": {
    "novaabase": {
      "url": "https://mcp.novaabase.com/v1",
      "headers": { "Authorization": "Bearer nb_live_•••" }
    }
  }
}

text

Tool                          Description
─────────────────────────────────────────────────────────
db.query                      Run a parameterised SQL query
db.list_tables                Enumerate tables and columns
storage.list                  List files in a bucket
storage.get_signed_url        Generate a 10-minute signed URL
auth.lookup_user              Find a user by id or email
functions.invoke              Invoke a deployed edge function
ai.embed                      Generate embeddings
realtime.broadcast            Publish a channel event

Build your own agent

typescript

import { experimental_createMCPClient } from "ai";

const mcp = await experimental_createMCPClient({
  transport: { type: "http", url: "https://mcp.novaabase.com/v1",
               headers: { Authorization: `Bearer ${process.env.NB_KEY}` } },
});

const tools = await mcp.tools();
// pass to streamText / generateText

Security

Every MCP request is signed with a scoped API key. Tools respect RLS as the calling user — agents can never read data the user couldn't.