atlassian-cli is the practical conclusion of Episode 04: most agent integrations don’t need to be MCP servers. A clean CLI with JSON output and the right subcommand boundaries gives the agent everything it needs, and gives you, the engineer, a tool you can debug from your terminal.
Letting an agent work with Jira typically means an MCP server: heavy schemas, every operation re-tokenised on every turn, brittle to API changes, hard to debug. For most workflows, this is overkill. Agents just need a small surface of well-shaped commands.
atlassian-cli ships a tight set of subcommands (`atl issue search`, `atl issue create`, `atl page get`), each returning JSON the agent can pipe directly into the next step. No schemas to load. No protocol layer. The agent invokes it like any other shell tool, the same way a human would.
In my own daily use, the CLI consistently beats the equivalent MCP server on token cost, latency, and reliability. Featured in Episode 04 of The Agentic Engineer as the canonical example of when a CLI is the better answer.