Just MCP Server

Just‑MCP – A Lightweight, Safe Bridge Between LLMs and Just

Just‑MCP is a production‑ready Model Context Protocol server that exposes the functionality of the Just command runner to large language models. By turning Justfile recipes into first‑class MCP resources, the server lets an AI assistant discover, validate, and execute build or workflow commands without ever having to parse the Justfile itself. This abstraction reduces the cognitive load on the model, keeping its context focused on higher‑level reasoning while delegating low‑level task orchestration to a proven tool.

The server solves the problem of command‑line hallucination that plagues raw bash access. Because every recipe is a declarative, idempotent operation defined in the Justfile, the assistant can ask for a command list (), pick an entry, and run it with precise parameters. The Justfile acts as a safety net: syntax errors are caught early, and any side effects are confined to the recipe’s defined steps. This mitigates accidental file deletions or network abuse that can occur when an LLM is given unrestricted shell access.

Key capabilities include:

• Recipe discovery – the server parses a Justfile and returns an inventory of available tasks, complete with names, descriptions, and parameter signatures.
• Execution orchestration – callers can trigger a recipe with optional arguments; the server runs Just, captures stdout/stderr, and returns structured results.
• Introspection – detailed metadata about each recipe (documentation strings, default values, environment variables) is exposed for richer prompting.
• Validation – Justfile syntax and semantic checks are performed before execution, ensuring that only valid commands reach the shell.
• Environment handling – .env files and variable expansion are supported, allowing recipes to run in the correct context without manual setup.

Real‑world scenarios that benefit from Just‑MCP include continuous integration pipelines where an AI assistant can trigger build steps, data science workflows that require reproducible preprocessing tasks, and automated documentation generation where the model can run on demand. Because Just is lightweight, the server incurs minimal overhead and works well with small‑to‑medium models (8k–32k context limits) that cannot afford the bandwidth of full shell access.

Integration into existing AI workflows is straightforward: an MCP‑enabled client simply queries the resource, receives a list of commands, and forwards user intents to the appropriate recipe. The server’s built‑in safety patterns—transparent logging, secondary model inspection hooks, and idempotent execution via git—provide a robust guardrail that keeps agents productive while preventing unintended side effects. In short, Just‑MCP turns the powerful but opaque world of Just into a clean, safe, and discoverable interface for intelligent assistants.