LUIS GUSTAVO PEIL DE OLIVEIRA

Posted on Jun 27

MCP is finally here: stop building REST boilerplate and start shipping

#mcp #ai #testing #programming

TL;DR: BugiaData MCP is live. Point Cursor, Claude, or Windsurf at a hosted SSE server, paste one config block, and ask your agent for relational fake data with valid foreign keys—no custom REST glue in your repo.

If you build with AI agents, you already know the loop:

You need realistic test data.
You wire up an HTTP client, auth headers, error handling, and JSON parsing.
You repeat that for every new tool, every new environment, every new IDE.

The Model Context Protocol was supposed to fix the integration tax. In practice, a lot of “MCP-ready” products still feel like REST APIs wearing a badge—same docs maze, same “here’s your API key, good luck.”

I built BugiaData to skip that layer for one job: generating fake data that actually behaves like a database—locales, multiple tables, foreign keys intact—directly from the chat in your editor.

What shipped

BugiaData now exposes a hosted remote MCP server (SSE) with four tools:

Tool	What it does
`ping`	Health check + tier hint
`whoami`	Quota and account info for your API key
`generate_data`	Single Faker-type batch (names, emails, etc.)
`generate_relational_data`	Multi-table JSON with `foreign_key` columns wired correctly

Same API key and monthly quota as the REST API and Studio. Free tier: 10,000 tokens/month.

Works today in Cursor, Claude Desktop, Claude Code, and Windsurf.

Why MCP beats another REST wrapper (for this use case)

REST is fine for humans. For agents, every endpoint is another chance to hallucinate paths, forget headers, or invent query params.

With MCP:

Your client discovers tools, not OpenAPI specs.
The agent maps natural language to structured arguments once.
You stay in the editor—no context switch to Postman or a seed script.

BugiaData’s differentiator isn’t “fake data exists.” It’s relational consistency: users and orders where orders.user_id actually references users.id, in pl_PL or pt_BR if you ask for it.

Flat Faker lists are easy. Graph-shaped test data is where projects stall. That’s the wedge.

Replace “call my REST API” with this (≈5 minutes)

1. Get an API key

2. Add the server to your client

Cursor — .cursor/mcp.json (project or global):

{
  "mcpServers": {
    "bugiadata": {
      "url": "https://mcp.bugiadata.com/sse",
      "headers": {
        "X-API-Key": "YOUR_API_KEY"
      }
    }
  }
}

Claude Desktop / Claude Code / Windsurf — same mcpServers block in your client config (claude_desktop_config.json, ~/.claude/mcp.json, or Windsurf MCP config).

Restart the client after saving.

3. Ask for data in plain language

Try something concrete:

Using BugiaData, generate 20 users and their orders. Orders must reference users.id. Use locale pt_BR.

Or a single-type batch:

Generate 10 company names and emails with locale en_US via BugiaData.

The agent should call generate_relational_data or generate_data and return JSON in chat—no shell scripts, no hand-rolled fetch.

See it before you sign up

The MCP landing page includes a live demo (rate-limited ping / sample generation) so you can watch JSON come back before creating an account:

👉 bugiadata.com/mcp

Full docs (REST + schema shape): bugiadata.com/docs

Listings (if you’re browsing, not building)

cursor.store: cursor.store/mcp/lgpoliveira/bugiadata-mcp
MCP Registry: io.github.lgpoliveira/bugiadata-mcp — search
Install snippets repo: github.com/lgpoliveira/bugiadata-mcp

What I’d love feedback on

If you’re already on MCP in production:

What would make you default to MCP over a one-off REST call for data tools?
Is relational schema generation the missing piece, or still overkill for your tests?

Drop a comment or open an issue on the docs repo. I’m especially interested in real agent prompts that broke—or surprised you—in a good way.

Links

DEV Community