just

just-mcp exposes the recipes defined in a project’s justfile as MCP tools that AI agents can discover, inspect, and run. Once started, the server watches the repository for changes to the justfile and dynamically updates the registry of tools so the AI can query what tasks exist, what parameters they accept, and how to invoke them. This enables approaches like natural-language prompts such as “run deploy to production” or “build the project” to translate into the corresponding justfile task invocations. If you enable vector search (optional), the MCP server can also provide semantic matching of tasks and parameters, helping agents find the most relevant automation even across multiple projects. Admin capabilities are exposed to refresh the tool registry or create tasks with AI assistance, further easing integration with automation workflows.

To use it, run the server (e.g., just-mcp) in the repository that contains your justfiles. Connect your MCP client or agent to the server name (commonly “just”) and use the client’s MCP capabilities to discover, ask for details, and execute tools. For example, an agent can query which tasks exist, inspect parameter descriptions parsed from comments, and trigger a specific task with the right arguments. The server supports real-time updates so changes to the justfile are reflected without restarting the client.

Prerequisites

• Rust toolchain (Rust 1.65+ recommended).
• Access to cargo and, optionally, cargo-binstall for faster binary installation.

Option A: Install a prebuilt binary

1. Install via cargo-binstall (fastest):

cargo binstall --git https://github.com/toolprint/just-mcp just-mcp

2. Or download a release tarball and install manually:

curl -L https://github.com/toolprint/just-mcp/releases/latest/download/just-mcp-$(uname -m)-$(uname -s).tar.gz | tar xz
sudo mv just-mcp /usr/local/bin/

Option B: Build from source

1. Clone the repository:

git clone https://github.com/toolprint/just-mcp.git
cd just-mcp

2. Build the binary with all features (including vector search if desired):

cargo install --path . --features all

Run the server

• From the project directory (containing the justfile(s)) start the server:

just-mcp

• The MCP server in this project exposes justfile recipes as tools. Ensure your repository contains valid justfiles and that the tasks you want AI to access are defined there.
• If you want the client to automatically monitor a specific workspace, you can pass a watch directory via MCP client configuration or command-line options (for example, using -w or equivalent, depending on your client).
• Optional vector search capability can be enabled to provide semantic task matching. This relies on local embeddings and does not require API keys.
• Security: the server validates inputs and exposes resource-limiting and timeouts; tune these settings if you’re running in a shared or CI environment.
• Common troubleshooting includes ensuring the justfile is discoverable from the server’s working directory, and that the AI client is configured to connect to the correct MCP server name (e.g., just).