devin-devkit

Unofficial community project. Not affiliated with, endorsed by, or sponsored by Cognition AI. Devin is a Cognition AI trademark.

Linux · Personal Windows · Corporate Windows · Compatibility · macOS status

Part of the awesome-devin ecosystem: the curated hub for the devin-* tools.

devin-devkit turns the public devin-powerups registry into selectable installation profiles. It is an installer/distribution tool, not a meta-package: each Python app gets its own isolated uv environment, and the Node bridge uses npm separately.

Source → tools → profiles

• devin-powerups/registry.json is the catalog and profile source of truth for the first-party devin-* tools, this separate DevKit distribution, the maintainer hub, and related artifacts.
• tools/export_devkit_manifest.py builds this repo's bundled manifest from the registry, including artifact, interface, audience, platform and environment metadata. The exporter rejects private or unregistered tools and verifies immutable Git commit references.
• src/devin_devkit/cli.py previews and installs the selected profile under an explicit execution environment. It does not edit Devin settings, inspect session databases, or install Slack/Obsidian/VM services.

Execution environments

One devin-devkit distribution supports three environments. Linux and Personal Windows use the extended runtime: local execution plus optional delegated workloads through Devin VM or QwenPaw when the selected tool supports them. Corporate Windows is intentionally self-contained: local execution only, with no VM, QwenPaw, Slack dependency, external compute, or workload delegation.

The installer defaults to linux on Linux and personal-windows on Windows. Corporate Windows is never inferred from the operating system; select it explicitly so restricted machines fail closed:

devin-devkit install qa --environment linux
devin-devkit install qa --environment personal-windows
devin-devkit install qa --environment corporate-windows

Environment entry points: Linux · Personal Windows · Corporate Windows.

Compatibility

The generated compatibility matrix is rendered from registry.json; unsupported entries carry an explicit registry reason rather than guessed compatibility. Entries marked Local only install without VM, QwenPaw, external compute, workload delegation, or required external integrations.

Profiles

QwenPaw is an optional add-on and is not installed by these profiles. macOS is planned, not tested.

Use

Install from PyPI (no runtime dependencies):

uv tool install devin-devkit

From a source checkout, use uv tool install . inside packages/devkit, or the subdirectory spec git+https://github.com/Icaro0310/devin-devkit.git#subdirectory=packages/devkit.

From a local checkout, use uv tool install .. Then inspect the available profiles and preview before applying:

devin-devkit profiles
devin-devkit list
devin-devkit install qa
devin-devkit install qa --apply
devin-devkit install full --apply

Install specs are pinned (PyPI versions or GitHub archive SHAs), so an install never drifts on its own. To move forward:

devin-devkit outdated          # which installed tools have newer pins
devin-devkit update            # preview the update plan
devin-devkit update --apply    # reinstall outdated tools
devin-devkit update --force --apply   # reinstall everything

outdated/update read the manifest published in this repository, which a weekly workflow re-exports from the maintainer registry, so the pins are never more than about a week behind upstream. Updates apply the same gates as install: the selected --environment, platform support, and prerequisite checks are enforced at plan time, and tools whose commands exist on PATH but were not installed by devkit (no uv receipt, no npm global entry) are reported as unmanaged and never force-reinstalled — remove them manually first if you want devkit to adopt them. A tool that was permitted and later became forbidden (policy or manifest changed) reports environment-blocked on outdated/update even when already at the pinned version — that is intentional, not a regression: the check exists to surface the violation, and current would hide it.

To run from a checkout without installing the DevKit itself:

uv run --project . python -m devin_devkit.cli install qa

The installer can also be run without installing: uvx devin-devkit ....

The manifest pins PyPI versions where available and GitHub source archives to immutable commit SHAs otherwise. Before it starts, the installer checks platform support, registry-declared environment compatibility, required managers, Node.js for devin-bridge, and existing command collisions. It never overwrites an existing command; commands already on PATH are reported as pre-existing but are not version-verified, and a partial collision blocks the profile before any installs run.

Runtime requirements

• Python tools: Python 3.10 or newer, installed in isolated environments managed by uv.
• devin-bridge: Node.js 20 or newer and npm. Installed from the npm registry as @icaro0310/devin-bridge.
• Most tools install from PyPI or HTTPS archives without Git. devin-skill-catalog currently installs from a git+https spec and requires git on PATH; Git is also needed for manual source checkouts such as devin-office.
• Supported environments: Linux, Personal Windows and Corporate Windows. Linux and Personal Windows may use optional delegated runtime; Corporate Windows rejects registry entries that require external runtime, delegation or external integrations. macOS is planned but not claimed as tested.

Adapters (MCP server, Devin skill, plugin)

devin_devkit.mcp_server exposes the read-only half of the CLI as MCP tools via the devin-devkit-mcp entry point (pip install 'devin-devkit[mcp]'). The write path stays CLI-only for a human.

adapters/ is a self-contained Devin plugin root (adapters/.devin-plugin/plugin.json + adapters/skills/devin-devkit/SKILL.md), installed with devin plugins install Icaro0310/devin-devkit#packages/devkit/adapters — it ships in the repo, not inside the wheel. The manifest launches the server through uvx --from 'devin-devkit[mcp]' devin-devkit-mcp, which resolves once the first PyPI release ships; until then run the source-installed devin-devkit-mcp directly or point a local manifest copy at the checkout (uvx --from './packages/devkit[mcp]' devin-devkit-mcp).

The same stdio server registers in other MCP hosts:

Zed — context_servers in settings.json:

{
  "context_servers": {
    "devin-devkit": { "command": { "path": "devin-devkit-mcp", "args": [] } }
  }
}

JetBrains — Settings → Tools → AI Assistant → Model Context Protocol → Add:

{ "mcpServers": { "devin-devkit": { "command": "devin-devkit-mcp" } } }

Antigravity — ~/.gemini/config/mcp_config.json:

{ "mcpServers": { "devin-devkit": { "command": "devin-devkit-mcp" } } }

Development and synchronization

uv run --extra dev pytest
python ../devin-powerups/tools/validate_registry.py
python ../devin-powerups/tools/export_devkit_manifest.py
python ../devin-powerups/tools/export_devkit_manifest.py --check
python ../devin-powerups/tools/render_catalog.py --check ../Icaro0310/README.md

The DevKit package has no runtime dependencies. New catalog entries and profiles belong in the registry; regenerate the manifest instead of hand-editing its generated copy. For local sibling development, install dependencies editably in a virtual environment (examples in the OS guides); keep relative file paths out of publishable package metadata.