Why SEP-1391 Might Be the Most Important Update You've Never Heard Of

If you’ve been anywhere near AI infrastructure or agentic systems lately, you’ve probably heard people talk about MCP — the Model Context Protocol.

Think of MCP as a common language that lets AI agents and tools talk to each other safely and consistently — kind of like how HTTP made the web interoperable.

So far, so good.

But until now, MCP had one big limitation: it assumed that everything happens instantly.

And if you’ve ever worked in advertising or marketing tech, you know nothing happens instantly.

Real Life Isn’t Instant

Most marketing operations take time — sometimes hours, sometimes days.

• You launch a campaign, but data ingestion takes hours.
• You run a lift study, but results come back next week.
• You request brand approval, but legal takes two days.
• You start creative rendering, and the videos are still transcoding tomorrow morning.

Every one of these steps is a long-running task — something that starts now, but finishes later, often after some human input or system dependency.

Until now, AI systems using MCP didn’t have a standard way to handle that.

They’d start a task and… well, either hang or disappear.

That’s exactly the problem SEP-1391 solves.

SEP-1391: Giving MCP “Patience”

SEP-1391 is a new proposal that adds long-running job support to MCP.

Here’s what that means in plain language:

Instead of expecting every job to finish right away, a tool can now say:

“I’ve got your request. It’s in progress. Here’s a token — check back later.”

That token acts like a tracking number for your AI task.

You can ask:

• What’s the status? (submitted, working, input_required, completed)
• Does it need human input before continuing?
• How long should I keep the result available (keepAlive)?

It’s exactly how real-world systems operate — finally expressed in the AI protocol layer.

What This Means for AdTech and Marketing

Let’s connect the dots.

The advertising ecosystem is a patchwork of asynchronous systems:

• DSPs optimizing bids
• SSPs clearing auctions
• Data clean rooms joining IDs
• Attribution engines back-filling conversions
• Compliance layers validating evidence bundles

Every one of these workflows involves waiting, resuming, or approving.

And every vendor currently handles it differently.

SEP-1391 gives all of them a shared language for asynchronous coordination.

Imagine:

• A clean room job can mark itself as input_required when a legal review is pending.
• A measurement system can expose a working status while calculating lift.
• A DSP optimization agent can check completed before moving to the next stage.

No custom integrations. No silent timeouts. No black-box jobs.

Just clear, auditable async handshakes between systems.

Compliance and Human-in-the-Loop Moments

One of the most powerful parts of SEP-1391 is that new middle state:

input_required.

That’s the protocol’s way of saying:

“Pause here — a human needs to approve this before we continue.”

In marketing, that’s everything:

• Approving a sensitive audience segment
• Reviewing an AI-generated creative
• Verifying a spend allocation
• Signing off on an attribution model

And because MCP tracks it all through one token, the entire chain — from action to approval to outcome — becomes traceable and auditable.

That’s a big step toward compliance-ready AI operations.

State Management: Who Remembers What

When you say “come back later,” someone has to remember where you left off.

That’s called state management.

The client (like an AI agent) keeps light state — job IDs, poll timers.

The tool (like a clean room or analytics engine) keeps heavy state — actual job progress, artifacts, and results.

The MCP protocol defines how both sides talk about that state.

That division of labor is what makes SEP-1391 scalable.

Each system only holds what it truly owns — and everyone speaks a shared language of status and results.

A More Interoperable Future

Let’s be honest: AdTech is famous for fragmentation.

Every system uses its own queue, its own job schema, its own status API.

SEP-1391 doesn’t eliminate that complexity overnight.

But it gives us a shared foundation — a way for every vendor, data partner, or AI agent to coordinate asynchronously without reinventing the wheel.

In short:

• MCP made AI systems interoperable.
• SEP-1391 makes them reliable.

The Bottom Line

If you build or operate in AdTech, SEP-1391 might sound like a small technical tweak — but it’s actually a massive step forward.

It means:

• Jobs won’t vanish when they take time
• Approvals fit naturally into automated flows
• Results can be audited after the fact
• Humans and agents can finally collaborate smoothly

The modern advertising stack is asynchronous by design.

Now, the AI layer finally speaks that language too.

In One Line

SEP-1391 gives AI agents the patience, memory, and accountability real AdTech has always needed.