---
name: eu-gov-data
description: "Retrieve EU public-affairs data through the eu-gov-data server: EU acts and case law (CELLAR), legislative procedures, MEPs and roll-call votes, Council documents, Commission consultations, registers and press, lobbying registers, EU funding, Eurostat and ECB statistics, specialist datasets and portal discovery. Use for a CELEX, ECLI or procedure reference, \"where does file X stand\", how an MEP voted, a register record; the general-purpose fallback. Not for: a full file brief (eu-file-tracker), group or country vote analysis (eu-vote-analysis), transposition (eu-transposition-monitor), a pipeline scan (eu-horizon-scan), a lobbying map (eu-influence-map), a funding map (eu-funding-map)."
metadata:
  stromy-client-summary: "Ask open questions of the EU record: acts and court cases, procedures, votes, Council and Commission documents, lobbying, funding, statistics and open-data portals."
---
<!--
  GENERATED FILE — DO NOT EDIT.
  Owner:       scripts/sync-mcp-skill-stubs.py (via sync-on-mcp-skill-change.yml)
  Source:      MCPs/eu-gov-data/skills/eu-gov-data/SKILL.md
  This workflow pushes DIRECT to this repo's main — a local edit here will be
  overwritten or rejected non-fast-forward. Edit the source, push, then:
    gh workflow run sync-on-mcp-skill-change.yml -R stromy-org/stromy-org
  Hand-authored skill? Set `_local: true` in frontmatter instead.
-->

# EU Gov Data (MCP-hosted skill)

This skill's full instructions are hosted on the `eu-gov-data` MCP server. Do not hardcode workflow logic locally — always fetch the live version from the MCP.

## Loading instructions

1. Read the main skill instructions:
   → call the `fs_read` tool on the `eu-gov-data` MCP with `path="skills/eu-gov-data/SKILL.md"`.

   **Read it to the end.** `fs_read` returns one page at a time. If the result's `next_offset_chars` is not null — or the returned text ends in a `<<< PARTIAL READ … >>>` block — the body is incomplete: call `fs_read` again with `offset_chars` set to that value and concatenate, repeating until it comes back null. Do **not** start work on a partial skill body. Hard rules and anti-patterns often sit in the final third, and a partial read fails silently — it looks like a complete skill.

2. Discover reference files (and any other skill assets), then read on demand:
   → call `fs_list` with `path="skills/eu-gov-data"` (and `path="skills/eu-gov-data/references"`),
   → call `fs_read` with `path="skills/eu-gov-data/references/<file>"`.

Follow the instructions returned by the MCP exactly.

## This MCP is the only correct path

Produce this skill's output **only** by following the live SKILL.md fetched above and calling the `eu-gov-data` MCP's own tools. Do **not** substitute a local or identically-named base skill from elsewhere, and do **not** invent your own output path. A locally-produced or unbranded artifact is **wrong output, not a fallback** — it bypasses the server-side brand and quality gates.

## If the `eu-gov-data` MCP is slow to respond

This server scales to zero to save cost, so the first call after an idle period wakes the container — typically ~10–30s, and up to ~1–2 min for a heavier image (media / browser tier). If `fs_read` or a tool errors with unavailable/timeout:

1. Tell the user the server is starting, then retry the same call — the call itself wakes the container.
2. Retry with a short backoff up to ~3 times.
3. Only if it is still unreachable after retries, STOP and report. Never downgrade to a local or base skill just to "get something out".