Vibe Coding #2 - Data API + AI Agent | OMIE Electricity Prices with MCP

Result

Ask “What was today’s electricity price?” → quarter-hourly price table and daily summary in seconds, directly from Claude Code via MCP.

Claude Code querying quarter-hourly OMIE prices via MCP: table with Spain €/MWh prices and daily summary

The problem

OMIE electricity market data is publicly available, but accessing it requires downloading CSV files manually
File formats are cryptic: semicolon-separated columns, market codes, one file per day
Every question means repeating the process: download, parse, calculate
No way to query this data in natural language

The solution

We started from the OMIE website, the Iberian electricity market operator, which publishes public day-ahead prices for Spain and Portugal.

OMIE electricity market clearing curve showing supply and demand

Architecture

We designed a three-layer system: raw OMIE data gets processed to Parquet, a FastAPI REST API exposes query endpoints, and an MCP server connects that API to the AI agent.

System architecture: User → AI Agent → MCP Server → FastAPI → OMIE Data

Building with Claude Code

Claude Code generated the entire project structure — data downloader, OMIE file parser, Pandas-powered aggregation service, REST API and MCP server. We worked with multiple agents: one in Visual Studio Code for initial scaffolding and another in NeoVim to continue development with shared context through the files.

Claude Code building the OMIE service in NeoVim terminal

The key insight: API First

The “aha” moment came when the MCP server was duplicating all the API logic. The fix: apply the API First pattern — build the API with FastAPI first, then mount it automatically as an MCP server with a single line:

# mcp_server.py - zero code duplication
mcp = FastMCP.from_fastapi(app=api_app, name="OMIE Energy Prices")

Any change to the API is automatically reflected in the MCP server.

Testing the system

With the API working, we tested price queries from a completely separate project — no access to the source code, only through the MCP server:

First MCP query returning OMIE prices from a separate project folder

Then we connected the server to Cloud.ai as a public connector, alongside integrations like Asana, Slack, or Figma:

OMIE data connector added to Cloud.ai alongside Asana, Slack, Figma

Interactive dashboard in the chat

The final step was building an MCP App in TypeScript that renders interactive charts directly inside Cloud.ai — no separate web app needed:

Interactive OMIE price dashboard rendered directly inside Cloud.ai chat

Reference

API endpoints

Endpoint	Description	Example
`GET /api/prices`	Prices by date range	`?start_date=2026-01-20&end_date=2026-01-27`
`GET /api/prices/aggregate`	Aggregation (mean, min, max, median)	`?operation=mean&column=price_spain`
`GET /api/prices/aggregate/grouped`	Grouped aggregation by date or hour	`?group_by=date&operation=mean`
`GET /api/dates`	Available dates	`?market_type=daily`
`POST /api/download`	Download OMIE data	`?start_date=2026-01-01&end_date=2026-01-31`

Project structure

main.pyCLI: download, process, server, mcp

src

omie_service.pyBusiness logic: download, parsing, aggregations

api.pyFastAPI REST endpoints (/api/*)

mcp_server.pyMCP server auto-generated from FastAPI

app.pyCombined server: /api/* + /mcp/*

mcp-appInteractive dashboard (TypeScript)

server.tsMCP server with visualization tools

srcUI rendered in Cloud.ai

data

rawOMIE CSV files

processedOptimized Parquet files

Stack

Python 3.12 / FastAPI / FastMCP / Claude Code / Pandas / UV / Cloudflare Tunnel

Resources

MCP Documentation - the standard protocol for connecting AI to external systems
OMIE - Iberian Market Operator - electricity price data source
OMIE Data - File access - downloadable CSV files
FastAPI - REST API framework
FastMCP - library to create MCP servers from FastAPI