aiaddons 0.2.0

Developers who want to use an MCP server, an agent skill, or a composite plugin with their AI coding agent currently have to configure everything by hand: hunt down the repository, decipher the target agent's config file path and format, manually edit JSON or YAML structures, handle tokens and credentials safely, and repeat all of this per agent when switching tools or working across multiple environments. aiaddons replaces manual configuration with a single command that installs, updates, removes, and synchronizes add-ons across whichever AI coding agents you use—transactionally, with full write-ahead rollback protection, and without touching config files by hand. Built to be the tool you reach for every time you want to add a new capability to your coding agent, the same way you'd reach for pip or npm.

Quick demo

$ aiaddons install github-mcp --dry-run

GitHub MCP Server
--------------------

Target:
  Claude Code
  Workspace

Plan:
  + Validate source
  + Prepare MCP configuration
  + Configure MCP server 'github-mcp' (@modelcontextprotocol/server-github) for Claude Code (workspace)
  + Inject key 'mcpServers.github-mcp' into agent configuration file '.claude.json'

No changes were made.

Why aiaddons

• Transactional installs with rollback: Every operation is logged to a write-ahead log (WAL). If validation, fetching, or configuration injection encounters an error, the operation rolls back to the previous known state so configuration files are never left in a broken or half-installed condition.
• Cross-agent from a single command: Add integrations to Claude Code, OpenAI Codex, Antigravity CLI, Cursor, or Hermes Agent without remembering agent-specific file paths, schema keys, or formatting quirks.
• Safe secrets and execution: Manifests declare required variables without storing credentials. Secrets are collected in-memory with masked terminal input, and background processes execute strictly via typed argument vectors rather than arbitrary shell scripts.
• Configuration drift detection: Inspects existing agent configurations and filesystem state before modifying or removing add-ons, warning you if files were modified externally since installation.
• Real verified registry: Manifests are validated against live package registries and genuine Git commit SHAs, preventing broken installs from fabricated packages or non-existent commits.

Installation

Install aiaddons from PyPI:

pip install aiaddons

You can also install aiaddons in an isolated environment using pipx or uv:

pipx install aiaddons
# or
uv tool install aiaddons

Prerequisites

• Python 3.11 or higher
• At least one supported AI coding agent installed

Supported agents

aiaddons detects and configures integrations across 5 AI coding agents:

To inspect your locally installed agents and detected configuration files, run:

aiaddons agents

Core commands

Registry

The built-in registry currently provides 35 verified integrations across three formats:

• Model Context Protocol (MCP) servers (17): Data connectors and external tools including github-mcp, postgres-mcp, sentry-mcp, stripe-mcp, supabase-mcp, brave-search-mcp, playwright-mcp, context7-mcp, chrome-devtools-mcp, heroku-mcp, browserstack-mcp, and qase-mcp.
• Agent skills (15): Prompt templates, instruction sets, and workflows centered around SKILL.md specifications, including caveman, vibesec-skill, skill-creator, and superpowers-*.
• Composite plugins (3): Multi-component packages combining MCP servers, skills, and CLI tools under unified configuration boundaries (superpowers, ponytail, python-lint-plugin).

The registry also includes preconfigured team stacks (dev-starter-stack.yaml, agent-behavior-stack.yaml) for zero-touch project onboarding.

Explore available add-ons using the CLI or the interactive terminal user interface:

aiaddons list
aiaddons search <keyword>
aiaddons info <addon-id>
aiaddons tui

For manifest specifications and schema documentation, see docs/REGISTRY.md. For engine internals and write-ahead log mechanics, see docs/ARCHITECTURE.md.

Contributing

Contributions are welcome. Before submitting pull requests or adding new registry manifests, please review CONTRIBUTING.md for local environment setup, test suite execution, and manifest verification requirements.

License

This project is licensed under the MIT License.