Why the compliance gate belongs in your AI workflow
More and more marketing teams draft, localise, and iterate campaigns with an AI assistant in the loop. The assistant writes variants, adapts a template for a new market, or generates a promotional banner. What it cannot do on its own is tell you whether the result is compliant — whether the safer-gambling notice survived, whether the bonus terms are prominent enough, whether a required logo is still in the rendered image. That judgement usually happens later, in a separate tool, by a separate person, long after the assistant has moved on.
The result is a gap. The creative is produced in one place and checked in another, and the check is the step most likely to be skipped under deadline. Closing that gap means making the compliance check something the assistant can run itself, as a normal step in its workflow — draft, check, read the result, fix, check again — all before a human is ever asked to sign off.
That is exactly what RuleRun’s MCP server does. It turns your compliance policy into a tool an AI client can call, so the gate moves to where the work already happens.
What MCP is, briefly
The Model Context Protocol(MCP) is an open standard for connecting AI assistants to external tools and data. A client that speaks MCP — Claude, Cursor, and a growing list of others — can discover the tools a server offers and call them during a task, passing arguments and reading structured results back. It is the same idea as a plugin or a function call, standardised so one integration works across many clients.
RuleRun runs an MCP server that exposes your compliance capability as a small set of tools. Any MCP client you connect it to can then trigger a run, read the verdict, and reference your rule sets — without a bespoke integration for each tool.
Same engine, new entry point
MCP is a front door, not a different product. A run triggered from an assistant goes through the identical pipeline as one triggered by email or the REST API: the email or image is rendered, checked against your rule set, and recorded with screenshot and PDF evidence. You are not getting a lighter check because it came from a chat window.
What the MCP server gives an agent
The tools map directly onto the things a workflow needs to do around a compliance check:
Trigger a run
The assistant submits an email’s HTML or a rendered image together with the rule set to check it against, and RuleRun starts a run — the same trigger the API and email paths use. The assistant gets back a run reference it can poll.
Read the verdict and breakdown
Once the run finishes, the assistant reads the overall pass / fail status and the per-rule breakdown — which rule failed, and the measurement behind it (for example, terms rendered at 48% of the headline size against a 90% threshold). That structured result is what lets the assistant explain the failure and propose a fix, rather than just reporting “failed.”
Reference your rule sets
The assistant can list the rule sets available to your account and check against the right one by name, so an iGaming promo is judged by your iGaming policy and a finance email by its own. The policy stays defined once, in RuleRun; the agent just points at it.
Connecting it
Setup is a one-time step per client, and it does not require writing any code:
- Create an API key in RuleRun. It identifies your account and scopes what the connection can do.
- Add RuleRun’s MCP server to your client’s MCP configuration, authenticating with that key. RuleRun’s own Connect screen gives you the exact values to paste.
- Confirm the tools appear. Once connected, the client lists RuleRun’s compliance tools, and you can ask it to run a check in plain language.
One connection, your whole policy
You do not configure rules in the client. Everything — rule sets, thresholds, required assets — stays in RuleRun, where your compliance team manages it. The MCP connection just lets an assistant use those rules. Change a policy in RuleRun and every connected assistant checks against the new version automatically.
A worked example
Imagine a marketer working with an assistant to adapt a welcome-bonus email for a new market. The assistant produces the localised creative, then — as a built-in step — calls RuleRun to check it against the market’s rule set. The run comes back FAIL: the translated “T&Cs apply” line rendered at 48% of the headline size, below the 90% prominence threshold.
Because the breakdown is structured, the assistant can say precisely what went wrong and why, adjust the template so the terms render larger, and run the check again — this time a pass, with an annotated screenshot and a PDF evidence record attached. Only then does the work reach a human approver, who is now reviewing a creative that has already cleared the mechanical checks, with the evidence in hand.
Draft, check, fix — before a person is ever asked to sign off.
The assistant handles the mechanical pass against your policy; your reviewers spend their attention on the judgement calls, working from a rendered email and an evidence record.
MCP, API, or email trigger — which to use
MCP is one of three ways to run a check, and they coexist. Choose by where the work lives:
Email trigger — zero code
Send a real campaign to a unique compliance address. Best for teams who want to validate before they wire anything up, or for an ESP step that just forwards a copy.
REST API — pipelines
Submit HTML or a rendered image programmatically. Best for CI/CD, approval tools, and campaign systems that already run their own automation on a fixed schedule or trigger.
MCP — AI workflows
Expose the check to an AI client so an assistant can run it conversationally, mid-task. Best when the creative is being produced or iterated with an assistant and you want the compliance loop to close in the same place, without a human switching tools to run it.
They share one source of truth
Whichever entry point you use, the rules, the rendered-email check, and the evidence record are the same. MCP does not fork your policy into a second place to maintain — it is another way to reach the one you already have.
Security and guardrails
A tool an assistant can call needs clear boundaries. The connection is authenticated with an API key scoped to your account, so an assistant only ever sees and checks against your own rule sets and runs — never another tenant’s. Revoke the key and the connection stops working immediately.
The tools are built around running checks and reading results, not around making irreversible changes on your behalf. And the model of use is unchanged from the rest of RuleRun: automation handles the objective, repeatable checks, and a human still owns the final sign-off and the judgement calls — tone, imagery, anything genuinely novel — that no rule should decide.
Still a first pass, not the final word
Letting an assistant run the check makes the gate faster and harder to skip; it does not remove the reviewer. Treat an MCP-run check the same way you would any automated pass: it clears the mechanical failures so the sign-off conversation is about what actually needs a person.
Getting started
If your team already works with an AI assistant on campaigns, connecting RuleRun over MCP is the shortest path to making compliance a step the assistant runs itself. Create an API key, add the MCP server to your client, and ask it to check your next draft against the right rule set — the policy, the rendered check, and the evidence come from the account you already have.
Key takeaways
- MCP lets an AI client call RuleRun directly, so the compliance gate lives inside the agent workflow instead of a separate tool.
- The tools: trigger a run, read the pass/fail verdict and per-rule breakdown, and reference your rule sets by name.
- A run triggered over MCP uses the same rendered-email check and produces the same screenshot + PDF evidence as any other run.
- Setup is a one-time, no-code connection authenticated with an account-scoped API key; rules stay managed in RuleRun.
- Use email triggers for zero-code, the REST API for pipelines, and MCP when creatives are produced or iterated with an assistant.
- MCP speeds the gate and makes it harder to skip — it does not replace the human sign-off on tone, imagery, and novel cases.