Model Context Protocol (MCP)

The Model Context Protocol (MCP) is a standard protocol that enables AI assistants to interact with development tools and access project context. Mise provides an MCP server that allows AI assistants to query information about your development environment.

Overview

When you run mise mcp, it starts a server that AI assistants can connect to and query information about your mise-managed development environment. The server communicates over stdin/stdout using JSON-RPC protocol.

::: warning The MCP feature is experimental and requires enabling experimental features with MISE_EXPERIMENTAL=1. :::

Usage

The MCP server is typically launched by AI assistants automatically, but you can also run it manually for testing:

bash

# Enable experimental features
export MISE_EXPERIMENTAL=1

# Start the MCP server (it will wait for JSON-RPC input on stdin)
mise mcp

Available Resources

The MCP server exposes the following read-only resources that AI assistants can query:

`mise://tools`

Lists all tools managed by mise in your project, including:

Tool names and versions
Installation status
Configuration source

`mise://tasks`

Shows all available mise tasks with:

Task names and descriptions
Task dependencies
Command definitions

`mise://env`

Displays environment variables defined in your mise configuration:

Variable names and values
Environment-specific overrides

`mise://config`

Provides information about mise configuration:

Active configuration files
Project root directory
Settings and preferences

Available Tools

The following tools are available for AI assistants to call:

`install_tool`

Install a specific tool version (not yet implemented)

`run_task`

Execute a mise task with optional arguments.

Parameters:

task (required, string): Name of the task to run
args (optional, array of strings): Arguments to pass to the task

Example:

json

{
  "task": "build",
  "args": ["--verbose"]
}

When the AI assistant calls this tool, it will execute the specified task and return the output, including stdout, stderr, and the exit status.

Integration with AI Assistants

Claude Desktop

To use mise with Claude Desktop, add the following to your Claude configuration file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json Linux: ~/.config/claude/claude_desktop_config.json

json

{
  "mcpServers": {
    "mise": {
      "command": "mise",
      "args": ["mcp"],
      "env": {
        "MISE_EXPERIMENTAL": "1"
      }
    }
  }
}

After adding this configuration and restarting Claude Desktop, the assistant will be able to:

Query your installed tools and versions
List available tasks in your project
Execute tasks directly (e.g., "run the build task")
Access environment variables from your mise configuration
View your mise configuration structure

Other AI Assistants

The MCP server uses standard JSON-RPC 2.0 over stdio, making it compatible with any AI assistant that supports the Model Context Protocol. Consult your AI assistant's documentation for specific integration instructions.

Examples

When integrated with an AI assistant, you can ask questions like:

"What version of Node.js is this project using?"
"List all the tasks available in this project"
"Run the build task"
"Execute the test task with verbose output"
"What environment variables are set by mise?"
"Show me the mise configuration for this project"

The AI assistant will query the MCP server to provide accurate, up-to-date information about your development environment and can execute tasks on your behalf.

Technical Details

The MCP server implementation can be found in src/cli/mcp.rs. It implements the ServerHandler trait from the rmcp crate to handle:

Resource listing and reading
Tool invocation (task execution)
JSON-RPC communication over stdio

For more information about the Model Context Protocol, visit the official MCP documentation.