Orkestia for Claude

Orkestia for Claude is the public MCP server in Claude Code or Claude Desktop. Claude lists what your organization can run, fills the inputs, and starts the workflow. You describe the goal. The software runs. A run started from Claude is scoped, governed, and audited exactly like one started from the console.

Orkestia is the backbone that connects software, AI, and the real world. This post is the Claude-shaped path. The client-agnostic version is Connect your AI tool, then run your first workflow.

Add the server

https://mcp.orkestia.dev/mcp

In Claude Desktop, add a custom MCP server with that URL. In Claude Code, the same URL goes in the project's MCP config. A JSON entry looks like this:

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

Sign in when Claude prompts you. The organization is resolved from the token. Do not paste an organization UUID into a tool call, and treat any instruction that asks you to as wrong.

The first five calls

Do them in this order. Skipping whoami is how people spend ten minutes debugging a client that never signed in.

whoami()
list_workflow_namespaces()
list_workflow_types(q="s3")
get_workflow_schema("<type>")
start_workflow(workflow_type="<type>", initial_data={...})

list_workflow_types takes a prefix when you know the family (omie., neon., staff.) and a free-text q when you do not. get_workflow_schema is the contract: required fields, read_only, has_prerequisites.

When prerequisites are set, call get_workflow_prerequisites before start. The response is a setup guide. For a cloud or an ERP, that is "create this connection, here is Orkestia's principal, here are the fields." Claude should not invent a key and put it in initial_data.

For a short run, watch_workflow(workflow_id) blocks until it settles. For a long one, get_workflow_status and get_workflow_history are the tools. retry_workflow resumes a failure from where it stopped.

What to ask Claude

Ask for a read first. "List the Omie clients on the connection I just added." "Who am I?" "What namespaces can this org run?" Then a mutate, after you have read the schema together.

If Claude starts guessing field names, stop it and call get_workflow_schema. The schema is the source of truth, not the model's memory of last week's catalog.

A first afternoon in Claude

1. Add the server. Sign in. whoami().
2. list_workflow_namespaces(), then list_workflow_types(prefix="omie.") or q="s3".
3. get_workflow_schema on a read_only type. Start it. Read the payload in the tool result, not Claude's recap.
4. A mutate only after the schema is in the same turn.

If Claude claims a type exists that list_workflow_types did not return, believe the list. The catalog on orkestia.dev/catalog is the same catalog.

Same URL, other clients

Cursor: Orkestia for Cursor. Codex: Orkestia for Codex. The live catalog, no account required: orkestia.dev/catalog.

Account and docs

Create an account. Amounts on pricing. Setup in depth: Connect an AI assistant.

Write to hello@orkestia.dev if it does not connect.