Build a lean skill layer from CLI help output

A developer swapped token-heavy Jira and GitLab MCPs for Markdown-based skills generated from the Atlassian and GitLab CLIs.

Done with Claude Code· Intermediate· Sep 2026

What they did
They decided the MCP setup was too expensive in tokens, so they moved to the Atlassian and GitLab command-line tools instead. Using Claude Code, they asked the agent to explore each tool’s help text recursively and dump the supported actions into separate Markdown documents. Those documents were turned into two small plugins, acli-skills and glab-skills, which let the agent reuse known commands instead of rediscovering them on every run.
How it went
The author says the new setup covers the same work as the MCPs while using much less token budget. No numbers were provided, and both repositories were published openly.
Tips
This works best for focused, repeatable integrations rather than broad, exploratory ones. The author still recommends MCPs for fast prototyping and internal tooling, especially when a task needs sampling or progress notices during long jobs.

Try it yourself with Claude Code

I use [Jira, GitLab, or another CLI] and want a lighter alternative to a token-heavy MCP. Inspect the main command and its subcommands with --help, then turn the useful actions into a Markdown skill with exact example commands for [open an issue, list merge requests, add a note, etc.]. Try it on one harmless read-only task first, then stop and ask before any step that would create, edit, remove, or submit something.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on news.ycombinator.com ↗

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.