Documentation contradicted itself: "DO NOT" section prohibited MCP_SERVER_URL and scripts, while "Sandbox Networking" section required them. Agents would be confused about correct approach.

Root Cause

Two distinct MCP contexts were conflated:

• Prompt tool calls: Use simple function names (get_calendar_events())
• Script execution: Requires proxy config (export MCP_SERVER_URL="http://host.docker.internal:80/mcp/riksdag-regering")

Changes

Scoped "DO NOT" warnings to prompt context only

• Renamed section: "DO NOT Try to Call MCP Manually From Prompts"
• Added explicit note: "Scripts ARE used by agentic workflows"
• Cross-referenced Sandbox Networking section

Example of corrected pattern:

# ✅ For scripts executed via bash
export MCP_SERVER_URL="http://host.docker.internal:80/mcp/riksdag-regering"
node scripts/generate-news-enhanced.js

// ✅ For MCP tool calls in prompts
get_calendar_events({ from: "2026-02-16", tom: "2026-02-16" })
search_voteringar({ rm: "2025/26", limit: 50 })

Files updated:

• .github/workflows/news-article-generator.md
• .github/workflows/news-evening-analysis.md
• .github/workflows/news-realtime-monitor.md

Both patterns are valid in their respective contexts. Documentation now reflects this.

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.