Making MCP Tool Calls Scriptable with mcp_cli (www.joshbeckman.org)

Anthropic/agent-style MCP tooling now has a simple, open‑source CLI: mcp_cli. It exposes MCP (Model Context Protocol) server “tools” (search notes, query logs, fetch docs, call Salesforce, etc.) as command‑line calls you can run with gem exec mcp_cli, returning structured JSON that you can pipe to jq or other Unix tools. The CLI is written for quick distribution via Ruby’s gem exec (no install beyond Ruby), detects MCP config used by major agent platforms (Claude, Cursor), and effectively wraps the underlying HTTP/cURL requests so you can call, test, and script MCP tools directly from bash. For AI/ML practitioners this is practical and consequential: mcp_cli lets you extract the final operational logic from an LLM conversation into reproducible scripts that can be scheduled, shared, or run without re‑invoking the model. That reduces model context usage, latency and token cost (e.g., streaming a long transcript through a pipe into a downstream MCP call rather than refeeding it to the model), improves privacy/efficiency, and enables lightweight “jigs” where agents compose multiple MCP operations into single pipelines. It’s a low‑infrastructure alternative to TypeScript execution examples—useful for prototyping, automation, and making agent-driven workflows deterministic and scriptable.