Koinoflow
Open source · free to self-host · MIT licensed

Turn the processes your team already runs into AI skills anyone can use

No Git, no code. Your ops, HR, and support teams publish governed AI skills in minutes. Claude, Cursor, ChatGPT, Gemini, and your own agents get only the skills they are allowed to use.

View on GitHub
MCP server live in 30 minutesTry the demoManaged hosting

app.koinoflow.com / skills / onboarding

Published

people · onboarding

New hire onboarding

v2.4.1

name: new-hire-onboarding

owner: [email protected]

description: Run the day-1 onboarding flow

Send welcome email with IT setup link
Schedule buddy intro within 48 hours
Provision accounts from role template
Day-1 checklist hand-off to manager
Calls (30d)

142

Clients

4

Health

92

Works with every major AI client that speaks MCP

ClaudeChatGPTCursorGeminiGitHub CopilotWindsurfClineContinueZedWarpany MCP-compatible client

Publish once. Every AI agent your team already uses follows the same governed process.

All product names, logos, and brands are property of their respective owners. Listed clients support the Model Context Protocol (MCP); Koinoflow is not affiliated with or endorsed by any of these companies.

The problem

Your AI agents are only as good as the context they receive

Three silent failures show up in every org rolling out AI: context is scattered, nobody owns it, and nobody can see what's being called.

Scattered context

Your SOPs, runbooks, and policies live across Confluence, Notion, Slack, and docs. AI agents grab whatever comes up first.

No ownership

Nobody is explicitly accountable for a process, so nobody updates it. Agents keep executing outdated guidance.

No observability

You cannot see which AI client called which skill, how often, or on which version. Risk and drift stay invisible.

The solution

Every skill, governed as a process.

A platform where every skill is governed as a versioned process with a named owner and an MCP interface, served to people and distributed to the right AI agents.

No Git, no code

Ops, HR, finance, support: anyone who owns a process can publish it. Guided visual editor, plain language, no YAML to learn.

Full version history

Every change is tracked, diffable, rollbackable. Agents always know which version of a process they consumed on any given call.

Delivery with agent scoping

Claude, Cursor, ChatGPT, Gemini, and headless agents read governed skills through MCP, with access controlled by role and agent.

Named ownership

Every process has a real person accountable for keeping it accurate. Staleness alerts fire when reviews are overdue.

Usage analytics

See exactly which skills were called, when, by which client, and by which agent. Adoption rates, health scores, per-client breakdowns.

Coming soon

Capture: skill discovery

Capture starts with Confluence and expands to additional document sources over time. AI surfaces existing skill candidates so you don't start from scratch.

Where skills become processes, and skills and processes are discovered instead of rewritten.

See five sample skills →

AI delivery layer

MCP server included. Zero integration work.

Any MCP-compatible AI client (Claude, Cursor, GPT, or your own agent) can receive governed skills from the same Koinoflow workspace. People and agents can see different skill sets, so a release bot, support bot, and finance bot do not receive the same instructions by default. Every use is logged so you have full observability.

  • Find the right approved skill before the agent acts
  • Load the current published version every time
  • Include supporting checklists when a process needs them
  • Route suggested improvements to the owner for review
  • Log which person or agent used which skill

Two steps. Done: (1) Publish a skill in the visual editor, no code. (2) Paste the MCP config below into Claude, Cursor, or any other client. For automation, create the agent in Koinoflow and deploy only the skills it should receive.

Governed delivery flow
Release agent asks for deploy guidance
  -> Koinoflow checks which skills this agent can use
  -> Current "Deploy to production" skill is delivered
  -> Usage is recorded against the release agent
  -> Process owner can review adoption and update the skill

MCP client config

{
  "mcpServers": {
    "koinoflow": {
      "url": "https://mcp.koinoflow.com/mcp"
    }
  }
}

Works with Claude Desktop, Cursor, Windsurf, Zed, Cline, Continue, and any other MCP client. Paste it into your client's config and you're connected.

Observability

The moat is the analytics dashboard

With full visibility into skill usage and staleness, you can prove the ROI of your AI initiatives and ensure every human client and automated agent is using the right context.

Example dashboard view
Preview

847×

"Onboarding" skill called last month

90 days

Since "Deploy" was last reviewed, alert fired

34

Health score on "Refund policy", action needed

Illustrative figures from a typical workspace after ~30 days of usage. Your real numbers appear in the dashboard as skill usage is logged.

Open source (MIT)
SOC 2 in progress
Encrypted at rest & in transit
Read-only connectors
Audit log export included
Security details →

Ready to give your AI agents governed skills and processes?

Koinoflow is open source and free to self-host. Your MCP server is live in 30 minutes.

View on GitHub

Open source (MIT) · free to self-host · managed hosting by Visionect