A home for
your agent skills.
Serve Markdown instructions to your agents through the Model Context Protocol (MCP). Run your own server today, or explore the planned hosted account flow.
Available now: local, self-hosted, Google sign-in and consented activity snapshots. Hosted MCP endpoints: coming soon.
Choose how
you run skillar.
Use the open-source tools on your own infrastructure, or explore the planned managed service. The choice is who runs the server, not whether your skills stay in Markdown.
| Setup | Local / self-hosted | Hosted account |
|---|---|---|
| Availability | Available in the repository | Google sign-in and activity dashboard available; MCP endpoint coming soon |
| Endpoint | Your local or self-hosted URL | Account-scoped MCP endpoint planned |
| Operations | Your config, deployment and credentials | Managed provisioning planned |
| Pricing | Open-source software; your infrastructure costs | Not announced |
Local CLI and client integrations: MIT. Registry server and Cloudflare Worker: AGPL-3.0-only.
Remote self-hosting requires your own configuration and release validation. A skill read is not permission to execute its instructions.
Your account.
Your activity.
Sign in with Google, connect a device, and choose what it may upload. Review its uploaded snapshots in the remote dashboard. Every device starts local only.
Signing in stores your Google ID, email, name, picture and sign-in times. No MCP endpoint or server is provisioned. Hosted pricing is not announced.
Sign in with Google
Google handles authentication. skillar stores your Google ID, email, name, picture and first and latest sign-in times, keeps your session in a signed, secure cookie, and retains no Google tokens.
Connect a device
Download a one-time grant from the dashboard and connect the local CLI. Then save separate upload consent for that device. Connecting alone uploads nothing.
Review activity
Inspect Sessions, Workspaces, Flow and Heatmap from uploaded snapshots. Views refresh when consented uploads change; capture times and report status show data freshness.
Your agent. Your workflow.
- Claude Code
- Codex
- Cursor
- OmO
- Gemini CLI
- Gajae Code
- OpenClaw
Local setup detects supported runtimes. Capture coverage and remote integration vary by client.
Write. Serve.
Put it to work.
Instructions stay in files you own. Agents find the right skill and read it when the task calls for it.
Start with a skill
A skill is a folder with a
SKILL.mdfile: a name, a description and the instructions your agent reads when a task calls for them.Give it a home
Index existing folders locally, or configure your own remote registry. Managed skill servers are coming soon.
Connect your agents
Search through
search_skills. Read the selected skill through MCP resources. Load the details on demand, not the whole library.
Just Markdown.
Actual know-how.
No new authoring language. A skill bundles instructions with the files that support them.
The local catalog indexes your folders in place. The registry exposes skills through MCP discovery and resource reads.
Read the registry docs---
name: skillar
description: Discover remote skills for a new task by searching its subject and intended outcome, then load only a relevant skill. Also discover native skill packs for requested combined workflows and create or update team skills/packs when requested. Searching does not activate a skill.
---
Use the Skillar MCP tools directly.Less setup.
More doing.
The local CLI can detect your runtimes, configure MCP, and install a background service. Preview its changes before you apply them.
Fresh install · macOS or Linux · Node.js 24.15+
Upgrading from gisool? Migrate firstnpm install -g skillarThen run skillar init. Installing over gisool collides on the gisool command, so migrate first.
# Preview without changing files
skillar init --dry-run
# Connect runtimes and install the background service
skillar init
# Inspect your skill activity
skillar openPrefer a foreground process? Instead of the service step above, run skillar init --no-service, then skillar serve.
Good questions.
What is an agent skill?
A folder containing a SKILL.md instruction file and optional supporting files. The name and description help an agent discover it; the body carries the actual instructions.
Can I use skillar today?
Yes. Install the local CLI with npm install -g skillar. Run skillar init --dry-run to preview setup changes. macOS or Linux and Node.js 24.15+ are required. If you are upgrading from gisool, read the migration guide first.
Is the hosted service live?
Google sign-in and the activity dashboard work. Sign-in stores your Google ID, email, name, picture and sign-in times. Activity uploads require a connected device and separate explicit consent. Managed MCP provisioning remains a preview; hosted pricing is not announced. The hosted accounts section describes the journey.
Does every MCP client work?
Skills require a host that can discover and read MCP resources. The repository includes runtime-specific integrations; compatibility and capture coverage vary. Check the protocol and compatibility guide for your client.
Where does my local data go?
The local daemon listens on loopback, uses a local bearer token, and stores indexed skills and captured session activity in local SQLite. Cloud uploads are off by default. Connect a device and save explicit upload consent in the dashboard to enable them. Your connected agent and model provider handle data separately.
Can I host a registry for my team?
Yes. The repository includes a Node.js registry and a Cloudflare Worker for authenticated distribution from private R2 releases. You provide infrastructure and credentials. See the deployment guide.