MCP Server
The AI Rule Engine can act as a Model Context Protocol (MCP) server, so an AI client — such as Claude, Cursor, or VS Code — can connect to your environment, run your rule sets, and read selected files and logs.
What it lets you do
Once connected, an AI client sees your rule sets as tools it can run on your behalf, can use the message prompts you publish, and can read (and, when allowed, write) the files and execution logs you have granted as resources. That lets an assistant trigger your automations, inspect what happened, and use the results as part of a larger task — all without you copying data back and forth by hand.
Enabling the server
The MCP server is off until an administrator turns it on for the environment. Once enabled, you create a connection token that a client uses to identify itself. As with an API key, the token value is shown only once when you create it, so copy it right away. Creating a token requires an administrator role, and deleting a token immediately cuts off any client using it.
Scoping a connection
A connection token is the security boundary: a client can only see and do what the token allows. The token form is organized into sections, and everything you grant is scoped to that one token.
Token details
You give the token a name and, optionally, an expiry date after which it stops working.
Rule set tools
Choose whether the token can run every rule set in the organization, or build an allow-list of specific rule sets. Each allowed rule set becomes one callable tool (see Rule sets as tools below).
File & log access
Four kinds of resource are granted independently, each at one of three levels — No access, Read only, or Read & write:
| Resource | Read grants | Read & write also grants |
|---|---|---|
| Storage files | List and read stored files. | Save files. An optional folder prefix (for example invoices/2025) can restrict access to one path. |
| Context files | List and read the context files a rule-set run can read. | Save context files. |
| Log entries | Read rule-set-run log entries. | Add log entries. |
| Log files | Read log attachments. | Attach log files. |
Read & write access is still confined to the token's allowed rule sets, its storage folder prefix, and its environment — a token can never reach beyond the scope you gave it.
Prompts
A single toggle enables prompt access for the token. When on, the token can use the message prompts published on any rule set that is already within its allowed rule sets (see Prompts).
Rule sets as tools
Every allowed rule set is exposed as one MCP tool. What that tool looks like to the client is controlled by the rule set's MCP tool definition, authored on the rule set's MCP tab:
- Inputs become the tool's parameters. Each input maps to a context key, carries a data type and description, and can be marked required.
- Outputs decide which context keys are returned to the caller after the run, as structured content.
- Behavior hints — mark a tool read-only or destructive so clients can warn the user before running it.
When a rule set invoked through an MCP tool calls another rule set with a Rule action, the nested run executes synchronously and the caller waits for it — so it counts toward the overall execution timeout. Keep deeply nested or long-running rule sets in mind when they are reachable from MCP.
Paused runs and form submission
When a rule set run pauses on a Human Intervention action, the MCP call returns the generated form link instead of the final result. The server exposes built-in tools that let the client read the pending form — its fields and any prefilled values — and submit it on your behalf, resuming the workflow. Submitting requires the action's Allow AI agents to submit setting to be on; the submission honors the link's expiry and use limits, and each accepted submission consumes a use.
Prompts
A rule set can also publish reusable prompts — named, templated message sequences authored on its Prompts tab, with arguments substituted as {{name}}. A client that connects with a prompt-enabled token can pick these prompts and fill in their arguments. A token sees a rule set's prompts only when that rule set is in its allow-list and prompt access is enabled.
Resources
Beyond tools and prompts, the granted files and logs are exposed as MCP resources the client can browse and read: stored files, context files, log attachments, and rule-set-run log entries (surfaced as runs, paged through a cursor). Context values are included only when the rule set marks them Expose to MCP server.
Connecting a client
The MCP screen helps you wire up popular clients: it offers one-click setup for tools like Cursor and VS Code, a ready-to-paste command for Claude, and the configuration details you need for any other MCP-capable client. In each case you point the client at your environment and give it the connection token.
Authorizing without pasting a token
Instead of pasting a token, a client can ask for access through a guided authorization flow: it sends you to a consent screen where you approve the connection and choose its scope, and the client receives its own access automatically. You stay in control of what it can reach, and you can review or revoke it at any time.
Approving and revoking connected clients is covered in OAuth & App Authorizations.