Orkestia for Cursor

Orkestia for Cursor puts the catalog next to the code. Cursor writes the change. Orkestia is how that change deploys, talks to a provider, opens a pull, or posts a title, with a schema, a permission, and a history.

Orkestia is the backbone that connects software, AI, and the real world. The MCP URL is the same one Claude and Codex use. This post is the Cursor-shaped path.

Add the server

Open Cursor Settings, MCP, and add a server with this URL:

https://mcp.orkestia.dev/mcp

A project-level mcp.json entry looks like this:

{
  "mcpServers": {
    "orkestia": {
      "url": "https://mcp.orkestia.dev/mcp"
    }
  }
}

Authenticate as your Orkestia organization when Cursor prompts you. The server scopes every call to that org on its own.

Confirm, discover, run

whoami()
list_workflow_namespaces()
get_workflow_schema("<type>")
start_workflow(workflow_type="<type>", initial_data={...})
get_workflow_status(workflow_id)
get_workflow_history(workflow_id)

Cursor is fast at filling JSON. That is useful after get_workflow_schema, and harmful before it. If a type is not read_only, read the schema in the same turn as the start. If has_prerequisites is true, run get_workflow_prerequisites and finish the connection in the console before you ask Cursor to mutate anything.

Retry and terminate are the same tools the console uses. retry_workflow(workflow_id) does not start a second run. It resumes the one you already have.

What this is for

A pull is open. You want a runner, a preview site, a ticket, or a comment on the pull, attributed to you. GitHub is a connection, covered in GitHub in the catalog. SPAD publishes the site on your cloud. Runners stay in your account. Staff is how a named actor does the same work under the org chart: Staff: actors that run the work.

You can browse the same catalog in the browser: orkestia.dev/catalog.

A first afternoon in Cursor

1. Add the MCP URL. Sign in. Call whoami().
2. Ask Cursor to list namespaces, then list types with q="github" or q="spad".
3. Open get_workflow_schema on a read-only type. Start it. Read the payload in the run, not the model's summary of the payload.
4. Only then pick a mutate: comment on a pull, enqueue a deploy. Paste the schema into the turn if Cursor starts inventing field names.

If a run fails, get_workflow_history is the document you send to a person, not a screenshot of the chat. retry_workflow resumes that id.

The generic table of status, watch, retry, and terminate is in Connect your AI tool, then run your first workflow.

Other clients

Orkestia for Claude, Orkestia for Codex, and the generic guide Connect your AI tool, then run your first workflow.

Create an account. Pricing. Docs: docs.orkestia.dev. Mail: hello@orkestia.dev.