Skip to content

Extensions ​

Crow's functionality can be extended with add-ons installed from the Extensions page. Each extension can include MCP tools, dashboard panels, Docker services, and AI skills.

Installing Extensions ​

  1. Go to Extensions in your Crow's Nest dashboard
  2. Browse available extensions by category
  3. Click Install and configure any required settings (API keys, passwords, etc.)
  4. The extension installs automatically — panels appear in the sidebar after a gateway restart

What Happens on Install ​

Depending on the extension type:

  • MCP Server — registers tools accessible to AI chat and Claude Code
  • Bundle (Docker) — pulls images, starts containers, opens firewall ports
  • Skill — adds behavioral prompts that guide AI responses
  • Panel — adds a dashboard page with web UI

Extensions with web UIs are automatically proxied through the gateway (no extra ports to open).

Removing Extensions ​

  1. Go to Extensions in the dashboard
  2. Find the installed extension and click Remove
  3. Containers are stopped, files are removed, firewall ports are closed

Local models live in the Model Catalog ​

Extensions no longer lists model bundles (bundles that serve an LLM, embedding, rerank or vision model). In the AI & Models group a single Local models card shows how many models are registered and running, and opens the Model Catalog page of the dashboard, where models are downloaded, adopted from disk, registered, started and stopped. A model bundle that is already installed stays under Installed with its usual actions until it is migrated. Installing a model bundle directly through the API is refused with model_bundle_moved.

GPU apps that are not model bundles, such as Ollama, LocalAI, Kokoro TTS or SDXL, still install from Extensions.

If you relied on the CUDA vLLM model bundles. vllm-cuda-embed, vllm-cuda-rerank and vllm-cuda-vision leave Browse. Their Model Catalog equivalents run on llama-server (Vulkan) on any GPU, NVIDIA included: qwen3-embedding-0.6b or embeddinggemma-300m for embeddings, qwen3-reranker-0.6b for reranking and qwen3-vl-4b-instruct for vision. An operator on a CUDA-only host who wants vLLM can still install the generic vllm app bundle.

Available Extensions ​

AI & Automation ​

ExtensionTypeDescription
Browser AutomationBundleStealth browser with VNC, form filling, scraping
OllamaBundleLocal AI models for embeddings and analysis
LocalAIBundleOpenAI-compatible local inference
Bot BuilderBuilt-inGmail, Discord, and Glasses agents — create, deploy, and manage agents from the dashboard. BYOAI auto-config, Messages integration, skill deployment.

Finance ​

ExtensionTypeDescription
Tax Filing AssistantMCP ServerFederal tax preparation with PDF ingestion

Media ​

ExtensionTypeDescription
Media HubMCP ServerRSS feeds, YouTube, podcasts, TTS, email digests
PodcastSkillPodcast publishing with iTunes RSS
SongbookSkillChordPro charts, transposition, setlists

Storage & Productivity ​

ExtensionTypeDescription
File Storage (MinIO)BundleS3-compatible file storage
NextcloudBundleFile sync via WebDAV
Obsidian VaultMCP ServerRead and search Obsidian notes

Smart Home & Gaming ​

ExtensionTypeDescription
Home AssistantMCP ServerControl lights, switches, sensors
RoMMBundleRetro game library and emulator

Social & Communication ​

ExtensionTypeDescription
Video & Audio CallsBundlePeer-to-peer video/audio calling with WebRTC

Infrastructure ​

ExtensionTypeDescription
TailscaleBundleVPN access from any device
Push Notifications (ntfy)BundleSelf-hosted push via ntfy app
TURN Server (coturn)BundleWebRTC relay for NAT traversal

Extension Web UIs ​

Some extensions provide web interfaces (VNC viewer, MinIO console, etc.). These are accessed in two ways:

Proxy Mode (default) ​

The extension UI is proxied through the Crow gateway at /proxy/<id>/. No extra ports or firewall rules needed. Works for simple apps (VNC/noVNC).

Example: /proxy/browser/vnc.html

Direct Mode ​

For SPA apps (React, Vue) that can't work behind a subpath proxy. The extension port is opened in the firewall and served via Tailscale HTTPS.

Example: https://your-machine.ts.net:9001/ (MinIO console)

This is configured automatically during installation — ports are opened and Tailscale HTTPS is set up.

For Developers ​

See the Creating Add-ons guide to build your own extensions.

Released under the MIT License.