CoreSkill is one home for the skills your company's AI follows: your sales process, your brand voice, your runbooks. Publish a skill once, each department gets exactly its own set, every person gets their own key, and you see who used what.
Product demo, current chapter: Build your org chart. The owner opens Departments from the sidebar, clicks New department, types Content, picks Marketing as the parent and clicks Create. The page confirms the department was created and the tree now shows Marketing with Content underneath.
Build your org chart
Mirror your company as a department tree, so each team can later get exactly its own skills.
Works with the AI apps your team already uses
Edit a skill in the browser or upload a new package. The very next question anyone asks reads the new version.
No files to email, nothing to re-download
Give a skill server to the whole company or to chosen departments, and every team underneath inherits it.
Move someone, their skills follow, same key
Every person holds their own key. Revoke it or let it expire and their next call is refused.
Expiry from 7 to 365 days, or none
Every call is logged with the person, department, skill, AI app and result, refused attempts included.
Filter, chart and export to CSV
The problem
Every company that hands its AI instructions out as attachments runs into the same six problems.
An honest comparison
Installing a skill file into each person's AI app works, and it wins in a few places. Here is where each approach is strong and where it is not.
you are one person or a very small team, skills rarely change, everyone uses the same AI app, or a skill depends on running bundled code.
several people or departments use the same skills, the content changes often, people use different AI apps, or you need to know who uses what and cut access the moment someone leaves.
How it works
Four steps. Each one has a clear owner, so nobody waits on IT.
Create your company's space and mirror your org chart inside it: Marketing, then Content and Social under it.
Up to 6 levels deep, each department with its own manager
Write them in the browser editor or upload a .zip or .skill package, then choose who sees them.
Whole company, chosen departments, or narrowed skill by skill
Invite by email link, then issue a personal key to one person, a whole department, or someone who has no account yet.
The key is emailed to the person it belongs to
They paste one address and their key into the AI app they already use. Their AI lists the skills it may read and follows them.
Every call shows up in your log straight away
Features
Built for a company with departments, managers and people coming and going, not for a shared folder.
Your real org chart, with permissions that follow it.
Group skills into servers and aim each one precisely.
One personal key per person, one address for the company.
Answers to real questions, not a wall of text.
A management connection for the owner's or admin's AI assistant.
Sensible defaults, and honest about the limits.
Use cases
What changes for each person once your skills live in one place.
Business owner
Result
The playbook everyone's AI follows is the one you approved
Department manager
Result
A process change reaches the whole team the same afternoon
Employee
Result
No more asking which file is the latest
Works with
MCP is the open standard that lets an AI assistant read from an outside system. CoreSkill speaks it over plain HTTPS, so it works with the AI app your team already uses, and the app gives a copy-ready snippet for each one.
{
"mcpServers": {
"coreskill-skills": {
"type": "http",
"url": "https://coreskill.app/api/mcp/biz/a3f1c9d2b7e4",
"headers": {
"Authorization": "Bearer cek_9f2a…"
}
}
}
}One address for the whole company
Each person pastes the company hub address and their own key. That is the whole setup.
Nothing to install, nothing to download, no add-on to keep up to date.
Security and control
The controls a company expects, enforced on every single request rather than trusted from a checkbox.
Only a SHA-256 hash mixed with a server-side secret is kept, compared in constant time. The real key is shown once.
The connection staff use has three read commands and no write or delete command at all, even if a key leaks.
Revoke a key, suspend a member or let a key expire: the next call is refused, and the refusal is logged.
Authenticator-app codes with single-use recovery codes, enforced for Google and Facebook sign-in too.
Block sign-ups from chosen email domains, including through Google and Facebook.
Two layers, by IP address and by key, stop floods before they reach the database.
Anything outside your business or your view simply answers not found, without confirming it exists.
Every admin change is recorded, and every call to the hub lands in a per-person usage log.
Off, Protected (no attachments, no raw reads for admin AI) or Locked (names and descriptions only), switched per business.
Skill bodies and files can be stored with AES-256-GCM, protecting backups, snapshots and stolen disks.
An AI can only follow a skill it can read, so whoever holds a working key can read what that key returns. Protection levels limit what leaves the server, and the per-person log shows who read what. Encryption at rest protects stored data, not against whoever operates the server. For full control, run it on your own infrastructure with Docker and PostgreSQL.
Pricing
Every business gets the full product at no cost and with no time limit. The limits below apply per business.
90-day trial
Every new business starts here automatically. No payment to begin.
No card, no countdown. Need higher limits? Contact the platform administrator.
FAQ
Still unsure? The help center walks through every screen.
CoreSkill is a skill hub for companies, served over MCP. A skill is the playbook you write for AI: a process, a brand voice, a runbook. You publish skills once, scope them to departments, and every employee's AI (Claude, ChatGPT, Cursor or Gemini CLI) reads the current version through your company hub with a personal key you can revoke at any time.
Yes, so that every key belongs to a real person in a real department, which is what makes the log worth reading. You can invite them by email, or issue a key to someone who has no account yet: they receive it by email and activate the account later. They install nothing.
They already have it. The very next question reads the new version, because there is no copy sitting on anybody's laptop to go stale.
No. What a key can see is worked out from the person's current department on every request, so moving them changes their skills automatically.
Only if you grant it. Positions, levels and one-off grants add read access to a department, a server or a single skill. They never grant the right to edit.
Any app that speaks MCP: Claude Desktop, claude.ai, Claude Code, Cursor, ChatGPT, Gemini CLI and others. The app shows a ready-made snippet for each one.
If their AI can read it, so can they, and no tool can honestly promise otherwise. What you do control: Protected mode stops attachments and raw reads by admin AI, Locked mode serves only names and descriptions, and the log records who read which skill.
In the PostgreSQL database of the deployment you use, optionally encrypted at rest. The log records which skill was read, never its content, and never the key itself. You can also run the whole thing on your own server with Docker.
Revoke their keys or remove them from the business and their next call is refused. Admin AI keys created by someone who is demoted, suspended or removed are revoked automatically.
Yes. Owners and admins can create a management key for their AI to create servers, edit skills, invite people and issue keys. Deletes need an explicit confirmation and every change is audited.
Each business has its own plan. New businesses start on a free trial; paid plans come in 3, 6, 12 and 24-month terms, and renewing adds time on top of what is left.
Create a business, publish your first skill and issue a key. Your team's AI can be reading it within the hour.
Cookies and privacy
We use cookies to measure how the site is used and to show relevant ads. You can turn them off now or at any time from Cookie settings. Cookie policy