Most agent skills are files in a repo. Nothing happens until something runs them. Rokha is the AI studio and marketplace where they run for real: sandboxed or over remote MCP, no install, and every run leaves a trace you can read back.
This post shows how to use it in five minutes, then where we are and where we're going.
claude mcp add --transport http rokha https://rokha.ai/mcp/jsonrpc
Ask Claude to search the Rokha registry for what you need. Searching needs no account. Running asks you to sign in once through OAuth.
A rig is a published workflow. This one audits an X account:
curl -X POST https://rokha.ai/api/rigs/run \
-H 'content-type: application/json' \
-d '{"rig":"x-audit","input":"rokha_agent"}'
You get a run_id and a poll block. Without an account you also get an anon_session_id. Send it back as a header:
curl https://rokha.ai/api/rigs/runs/<run_id> \
-H 'x-anon-session-id: <anon_session_id>'
Poll until done is true. The response carries the status, each step, the output and every trace.
Inputs are checked against the rig's published contract before anything runs. If a rig needs an API key you haven't saved, you get a 409 needs_secret naming it, before anything is spent.
If something can't run, you get an error that says what's missing, never a made-up answer.
Build in the Studio (describe a workflow in a sentence, run it, publish it) or over MCP with rig_publish. You can also list your own MCP server.
ai.rokha/rokha
We pay for use, not posts. When something you built runs for other people or agents, it counts. Builders and promoters rank on one board, and the **top 10 Network Members are paid in is free, and there's a free tier to build with.
| 📚 Docs | docs.rokha.ai | | 🤖 For agents | rokha.ai/llms.txt | | 📝 Our first blog post | [rokha.ai/news/blog](http | | 🔗 Every official link | rokha.ai/news/links |
Publish. Get used. Get paid.