Kiwi

Kiwi MCP is a unified server that consolidates directives, scripts, and knowledge management into four consistent tools. It provides a straightforward way to search, load from a registry, execute actions, and get guidance, so you can manage complex AI workflows efficiently from a single place.

How to use

To work with kiwi mcp you interact with four core tools through your MCP client: search, load, execute, and help. You can search for directives, scripts, or knowledge entries across local storage and the registry. Load items from the registry into your local workspace or project space. Use execute to run directives, publish items to the registry, delete items, or create and link new items. If you ever need quick usage guidance or troubleshooting tips, use help to get targeted information.

Typical usage patterns include:

• Searching for a directive related to authentication workflows to understand recommended steps
• Loading a script from the registry into your project workspace for execution
• Running a directive to bootstrap a project with a predefined setup
• Publishing a knowledge entry that documents best practices and versions for your team

How to install

Prerequisites you need before installing Kiwi MCP:
- Python 3.11 or newer
- pip for Python package installation (or poetry if you prefer)
- Access to a PostgreSQL-compatible registry backend (Supabase) if you plan to use the registry features

Install steps you should follow locally:

1. Clone the Kiwi MCP project to your machine.
2. Install development dependencies so you can run and test the server.
3. Start the MCP server and connect your MCP client to it.

# Step 1: Prepare your environment and install dev dependencies
# Ensure Python >= 3.11 is installed

# Step 2: Install the package in editable mode with development extras
pip install -e ".[dev]"

# Step 3: Run the MCP server locally
python -m kiwi_mcp.server

Configuration and startup

Before starting the server, you may provide registry connection details if you plan to use the registry features. The typical setup uses environment variables for the registry and user space, which you can set in your shell or in a configuration file used by your MCP client.

# Typical environment setup for registry usage and user space
export SUPABASE_URL="https://your-project.supabase.co"
export SUPABASE_SECRET_KEY="your-secret-key"
export USER_SPACE="~/.ai"   # optional, defaults to ~/.ai if not provided
export LOG_LEVEL=INFO

# Start the server
python -m kiwi_mcp.server

Additional content

Notes on usage and maintenance:

• The server exposes four tools with unified interfaces for consistent operations across directives, scripts, and knowledge entries.
• You can run, publish, delete, create, update, and link items using the execute tool, with behavior tailored to the item type.
• For troubleshooting, ensure your registry connection is reachable if you plan to use registry-backed features; local operations remain available without registry access.

Available tools

search

Find directives, scripts, or knowledge entries across local storage or the registry using natural language queries.

load

Download items from the registry to local storage or project space, preparing them for use.

execute

Perform actions on items such as run, publish, delete, create, update, or link, with behavior specific to item type.

help

Provide usage guidance, examples, and troubleshooting tips to help you work efficiently with the MCP server.