using-heavy-mcps_skill

Overview

This skill shows how to use mcporter and jq to avoid token bloat when calling heavy MCPs (e.g., Sanity, Brain vault). It describes a practical pattern: run MCPs outside of chat, filter and compact results, then paste only the essential JSON/text back into the model. The goal is repeatable, low-token workflows for production agent interactions.

How this skill works

You run mcporter commands from the shell (bunx mcporter call ...) and pipe the JSON output into jq or use mcporter CallResult helpers to extract text/markdown/JSON. The filtered, minimal output is what you paste into chat or embed in automation rules. You can save one-liners as aliases or include them in Cursor rules for repeatable, efficient fetches.

When to use it

• MCP responses are very large (full documents, long file contents)
• You repeatedly query the same data but only need specific fields
• You need to chain multiple MCP calls without bloating context
• You want a compact list (titles, ids, short metadata) instead of full items
• You’ll reuse the filtered output in rules, scripts, or agent prompts

Best practices

• Prefer jq for flexible, precise filtering; use CallResult helpers for simple text/markdown extraction
• Test the raw mcporter call first to understand structure, then add filtering
• Use --output json and structured error handling to detect failures without loading traces
• Set sensible timeouts (--timeout) and verify auth before expensive queries
• Save common queries as shell aliases or reusable Cursor commands

Example use cases

• Fetch project list from Sanity and return [{title,slug}] for prompts
• Search Brain vault and return top 5 note titles and paths for summarization
• Embed a compact portfolio-list command into a Cursor rule for agent context
• Run a mcporter call and jq-filter to provide only ids and statuses for bulk workflows
• Alias frequent queries in .zshrc to avoid retyping long commands