Skip to installation

Model Context Protocol

Give your AI agent the whole workspace.

Connect once. Your agent can call published flows, read the grammar and workspace context, and build and test drafts with the permissions you choose.

https://api.arborule.com/mcp

Remote HTTP server · OAuth or API key · no package to install

Connected workspacestreamable HTTP
Your AI agentasks, builds, calls
ArboRule MCP/mcp
toolsPublished flowsAuthoringTesting
contextgrammar://workspace://

One URL. The tools change with your workspace and permissions.

MCP is the shared language between an AI client and the tools and context it may use.

Without MCPCopy data into chat. Translate the answer back by hand.
With ArboRule MCPThe agent works with the real flows in the workspace.

What it exposes

One connection for the full flow lifecycle.

Call published flows

Your agent lists the workspace's published Decision Flows, reads what each one decides and what it wants, then runs one by slug. It receives the outcome and trace.

list_flows · run_flow · get_decision

Build and test drafts

An agent can read the flow grammar, create a draft, validate it, patch it, and run a real Sandbox test before anything goes live.

get_grammar · push_draft · test_flow

Publish with permission

Publishing is a separate permission. Give an agent authoring access without giving it live traffic or the right to publish.

publish_version

Connect your agent

Choose where your agent runs.

Hosted connectors can open a secure sign-in window, so they use OAuth. Local agents usually read a configuration file, so they use an API key. Both connect to the same server.

Nothing to install. Add the remote server by URL and let OAuth handle the credential.

  1. 1
    Open your connector settings

    Choose the option to add a custom MCP server or connector.

  2. 2
    Paste the ArboRule server URL
    https://api.arborule.com/mcp
  3. 3
    Sign in and choose a workspace

    ArboRule shows the permissions being requested before you approve access.

Permissions

Give the agent a job, not a master key.

Capabilities are separate, so the safe default can stay small. In particular, authoring never implies Live or Publish.

PermissionOAuth scopeWhat it grants
Readflows:readSee tools, flows, grammar and workspace resources.
Sandboxflows:rehearseTest drafts without touching live traffic.
Liveflows:runCall published flows against live data.
Authoringflows:authorCreate and edit drafts, but never publish them.
Publishflows:publishPublish a version. Kept separate on purpose.
Historydecisions:readRead decisions already made, with their inputs and outputs.
Recordsdata:writeWrite this workspace's Entities, Events and Cases.

Common questions

Before you connect an agent.

Do I need to install an npm or Python package?

No. ArboRule is a remote MCP server over HTTP. A hosted connector needs only the server URL; a local agent needs the same URL and an API key in its MCP configuration.

Which AI clients can connect?

Any client that supports remote HTTP MCP servers can connect. Hosted products such as ChatGPT and Claude use OAuth; local agents such as Claude Code and Cursor can send an API key in the Authorization header.

Can an agent edit a flow without running live decisions?

Yes. Authoring, Sandbox, Live and Publish are separate permissions. A key or OAuth grant can create and test drafts while having no access to live traffic and no permission to publish.

What does the agent see when it connects?

A fixed set of tools. It lists this workspace's published flows and runs one by slug, and it also gets the tools that build them, the tools that write this workspace's own Entities, Events and Cases, the reads that show what a flow actually did, the flow grammar and examples, workspace resources, and an authoring prompt. Publishing a flow does not change the tool list — it changes what the list of flows returns.

Where do I create an API key?

Open your ArboRule workspace, go to Settings, then API keys, and create a key with only the permissions the agent needs. Copy it when it is created because the full value is shown once.

Can I revoke access later?

Yes. Deactivate an API key from Settings at any time, or revoke a hosted connector's OAuth access. Requests using that credential stop authenticating immediately.

Start with a draft

Connect an agent without giving it production.

Create a workspace, add an MCP connection with Read, Sandbox and Authoring, and let the agent leave its first flow as a tested draft.

Read the flow documentation