Which plans include MCP?

Every plan. Reading is free. AI work the agent triggers in Orbox draws credits like any other work in the workspace.

Yes. Claude, Codex and ChatGPT sign in through the browser, no token to copy. A connection lasts 90 days and you can revoke it anytime in Settings → Apps and tokens. Scripts and other clients use API tokens.

Does the agent get access to my whole account?

It gets what you approve: read-only or read/write, in the workspace you’re signed in to. It acts as you, so it sees only what you can see.

Can a Partner use MCP in a client’s workspace?

Yes. The client invites the Partner as a member, and the Partner connects their agent while signed in to that workspace. Revoking the membership revokes access.

MCP server

Your AI agent, working on your processes.

Orbox runs a hosted MCP server. Connect Claude, Codex or ChatGPT by signing in through the browser, and the agent reads your processes step by step, or maps and edits them for you.

Set it up For developers

Server URL

https://orbox.dev/api/mcp

claude

Read the “Invoice processing” process and automate the next step in n8n.

How it works

Sign in, no token to copy: Claude, Codex and ChatGPT connect over OAuth. You sign in with your Orbox account in the browser and approve the connection.
You choose the access: Read-only, or read/write if the agent should map and edit processes. Nothing is pre-selected. The agent acts as you and sees only what you see.
Under your control: A connection lasts 90 days. You see and revoke it in Orbox under Settings → Apps and tokens, and organization owners see and revoke every member’s connections. Every change lands in the version history.
Scripts and CI too: For automation and clients without OAuth, create a personal API token. Read-only or read/write, valid 1 to 365 days.

Set it up

Three steps: connect, install the skill, try it.

1. Connect your agent

Claude Code

terminal

claude mcp add --transport http --scope user orbox https://orbox.dev/api/mcp

Then in Claude Code run /mcp, pick orbox → Authenticate and sign in in the browser. Check with claude mcp list.

Claude (web and Desktop)

Customize → Connectors → + → Add custom connector
Name: Orbox. URL: https://orbox.dev/api/mcp
Leave the advanced OAuth fields empty, then Add → Connect

One setup works in claude.ai and in Claude Desktop. On Team and Enterprise plans the owner adds the connector first, then each member connects.

Step-by-step guide for Claude

Codex

terminal

codex mcp add orbox --url https://orbox.dev/api/mcp
codex mcp login orbox

The second command opens the Orbox sign-in. If Codex stops authenticating after a while, run codex mcp login orbox again.

ChatGPT

Settings → Security and login → turn on Developer mode
Settings → Plugins (chatgpt.com/plugins) → + → create an app. Name: Orbox. URL: https://orbox.dev/api/mcp. Authentication: OAuth
Sign in to Orbox. In a chat pick + → Developer mode → Orbox

Web only. On Business, Enterprise and Edu the workspace admin enables developer mode. According to OpenAI, write actions are currently available on Business, Enterprise and Edu; other plans may get read access only. ChatGPT asks you to confirm every write.

Cursor, opencode and other clients connect with an API token. See API tokens

On the Orbox consent screen

Check that the redirect host is the app you started from: localhost for Claude Code and Codex, claude.ai for Claude, chatgpt.com for ChatGPT. Never approve a host you don’t recognize.
Pick Read and write to map and edit processes, or Read only just to read. Then click Allow.

2. Install the orbox-process skill

The MCP gives the agent the tools. The skill gives it the method: how to interview you, split a process into steps, resolve roles and tools, and review before submitting. Without it the agent still works, but the processes come out much worse.

Claude Code

terminal

claude plugin marketplace add ordexai/orbox-skills
claude plugin install orbox@orbox

Update later with claude plugin marketplace update orbox, then claude plugin update orbox@orbox.

Claude (web and Desktop)

Download orbox-process.zip
Customize → Skills → + → Create skill → Upload a skill
Turn the skill on

Needs Code execution and file creation enabled in Settings → Capabilities. On Team and Enterprise the owner can share the skill with the whole organization.

Download orbox-process.zip

Codex, Cursor and others

terminal

npx skills add ordexai/orbox-skills -g

Installs the skill globally for the agents detected on your machine. Update later with npx skills update orbox-process.

Start a new session after installing, the agent loads skills at the start. The skill is public and licensed for use with Orbox: github.com/ordexai/orbox-skills

3. Try it

In a new conversation, check the connection first, then start working:

What’s in my Orbox workspace? Show me the processes and the entity catalog.
Map our process for approving a supplier invoice.
Show me what needs fixing in the “Invoice processing” process.

After submitting, a process goes through a quick AI check first (waiting_for_processing). The agent waits for it with get_process_status, then hands you a link to the draft in Orbox. A human verifies and approves it there.

API tokens for scripts, CI and other clients

For automation and for clients that don’t support OAuth. Same server over Streamable HTTP, the token goes in an Authorization: Bearer header.

1.
In Orbox open Settings → Apps and tokens and create a token. Name it after the device or tool.
2.
Pick read-only or read/write and the expiry: 90 days by default, 1 to 365.
3.
Copy the secret. It’s shown only once. Keep it in an environment variable, never in a repository, a URL or a shared project config.

Environment variable

~/.zshrc

export ORBOX_API_TOKEN='paste-token-here'

Cursor

~/.cursor/mcp.json

{
  "mcpServers": {
    "orbox": {
      "url": "https://orbox.dev/api/mcp",
      "headers": { "Authorization": "Bearer ${env:ORBOX_API_TOKEN}" }
    }
  }
}

Works the same in any client that supports remote MCP with custom headers.

opencode

~/.config/opencode/opencode.json

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "orbox": {
      "type": "remote",
      "url": "https://orbox.dev/api/mcp",
      "enabled": true,
      "oauth": false,
      "headers": { "Authorization": "Bearer {env:ORBOX_API_TOKEN}" }
    }
  }
}

"oauth": false stops opencode from trying the OAuth flow.

Claude Code with a token

terminal

claude mcp add --transport http --scope user orbox https://orbox.dev/api/mcp \
  --header "Authorization: Bearer $ORBOX_API_TOKEN"

For CI or a machine without a browser. Keep --header last.

Codex with a token

terminal

codex mcp add orbox --url https://orbox.dev/api/mcp --bearer-token-env-var ORBOX_API_TOKEN

Codex reads the variable at startup, so export it in the shell that launches codex.

REST API

For integrations from your own systems there’s a REST API v1 with the same tokens. Interactive docs: orbox.dev/api/v1/docs and the OpenAPI spec: openapi.json

What the agent can do

Read tools work with any connection. Write tools need read/write access.

Area	Tools	Access
Orientation	get_workspace_overview, search_processes, search_entities, list_entity_types, list_entities	Read
Reading	get_process, get_entity, get_process_status, get_process_history, get_process_comments, get_lint_status	Read
Processes	create_draft_process, update_process_meta, apply_process_batch, submit_process, archive_process	Read/write
Steps and flow	create_step, update_step, delete_step, reorder_steps, create_flow, delete_flow	Read/write
Entities	create_entity, update_entity, apply_entity_batch	Read/write
Comments	add_process_comment	Read/write

The agent can’t approve entity proposals, change entity types or fields, or delete anything permanently. That stays with people in Orbox.

When something doesn’t work

The agent doesn’t see the Orbox tools: Start a new conversation or restart the app. Check that the connection is on: /mcp in Claude Code, the connector toggle in Claude.
Sign-in fails or the consent page shows an error: Start the connection again from the client. Each attempt needs a fresh sign-in, an old link doesn’t work.
FORBIDDEN on a write: The connection is read-only. Reconnect and pick read/write, or create a read/write token.
STALE_REVISION: Someone changed the process in the meantime. The agent reloads it with get_process and repeats the change.
TOO_MANY_REQUESTS: Too many calls at once. Wait a moment, then batch the changes with apply_process_batch.
PROCESS_NOT_EDITABLE: The process is waiting for review or already approved. The agent creates a new draft version.

What to ask

Read the “Invoice processing” process and automate the next step in n8n.
Interview me about our onboarding and map it as a new process draft.
What changed in “Purchase approvals” since last week, and what do I need to rebuild?

Questions

Which plans include MCP?
Every plan. Reading is free. AI work the agent triggers in Orbox draws credits like any other work in the workspace.
Is there OAuth?
Yes. Claude, Codex and ChatGPT sign in through the browser, no token to copy. A connection lasts 90 days and you can revoke it anytime in Settings → Apps and tokens. Scripts and other clients use API tokens.
Does the agent get access to my whole account?
It gets what you approve: read-only or read/write, in the workspace you’re signed in to. It acts as you, so it sees only what you can see.
Can a Partner use MCP in a client’s workspace?
Yes. The client invites the Partner as a member, and the Partner connects their agent while signed in to that workspace. Revoking the membership revokes access.
Something doesn’t work?
Email us with the client you use and the error message. We reply by email.