Orkestia now runs on Codex

Orkestia now runs on Codex. The catalog you open in the console, typed workflows, connections, identity, is callable from OpenAI Codex over MCP. You describe the goal. The software runs. The run is the same object a person would start from the console, with the same permission and the same history.

This is the announcement. The client-specific walkthrough is Orkestia for Codex. The shortest path from a fresh account to a finished run, for any MCP client, is Connect your AI tool, then run your first workflow.

Why this pairing exists

Codex is already in the loop where code is written. Orkestia is the loop where that work reaches a cloud account, an ERP, a payment provider, GitHub, or Staff. Connecting the two is one paste, not a second product.

A run started from Codex is not a side door. It passes through the same access control as the console. Credentials stay on the connection. Codex sees types and schemas, not keys.

The URL

https://mcp.orkestia.dev/mcp

Add it as a streamable HTTP MCP server in Codex. Sign in when the client prompts you. The server resolves the organization from the token. You do not pass an organization id by hand.

Then, in this order:

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

whoami tells you the connection is real and which identity the next run will be recorded under. get_workflow_schema is the one that matters before anything moves: required fields, and whether the type is read-only. If has_prerequisites is true, call get_workflow_prerequisites before you start.

What you can actually run

Whatever your organization can run in the console. Browse it without an account at orkestia.dev/catalog. Typical first runs:

• A read-only list, to see a payload come back.
• A connection test, after you stored a provider once.
• A deploy or a ticket, once you have read the schema.

Claude and Cursor take the same URL: Orkestia for Claude, Orkestia for Cursor.

A first run that cannot hurt much

Pick a read_only type from the catalog. Call get_workflow_schema on it. Start it with the fields the schema requires, usually a connection_uuid you already created. Watch it. Open the history.

That is the whole loop. The next run can be a mutate. The history from the read is how you know Codex is talking to your org, not to a cached demo.

If Codex offers to paste a cloud key into initial_data, refuse. The prerequisite guide is the path that stores the key on the connection.

Account and price

Create an account and subscribe. There is one published plan and every customer pays the same rates. The amounts live on pricing.

What to read next

• Orkestia for Codex, the setup in Codex itself.
• Connect an AI assistant.
• The documentation.

If you get stuck, write to hello@orkestia.dev.