For the last decade, the revenue tech stack has run on tabs. A rep works a LinkedIn profile in one tab, flips to Salesforce in another, opens an enrichment extension to pull an email address, pastes it into a sequencer, and logs the activity by hand. Multiply that by every seller, every day, and you get the real hidden cost of GTM software: not the licenses, but the human glue holding them together.

That glue is what's being replaced right now — not by another point tool, but by a protocol. It's called MCP, and if you own revenue systems, it's worth understanding well before your stack vendors force the conversation on you.

What MCP actually is

Model Context Protocol is an open standard, released by Anthropic in late 2024, that defines a common way for AI applications to talk to external systems — CRMs, data warehouses, enrichment platforms, calendars, ticketing systems, internal APIs. It runs on a JSON-RPC interface with three roles: a host, the AI application a person or agent interacts with; a client inside that host that speaks the protocol; and a server that exposes one specific tool or data source.

The comparison people reach for is USB-C. Before it existed, every device needed its own cable and its own port. After it existed, one standard plug worked everywhere. Before MCP, every AI-to-tool connection was a bespoke integration built by a developer, one pair at a time. With MCP, you build the connector once, expose it as a server, and any compatible agent can use it without custom code.

That sounds like plumbing. In GTM, plumbing is the whole business.

Why it matters for the team that owns your stack

The GTM engineering function exists because revenue teams accumulate dozens of disconnected systems — CRM, enrichment, intent data, dialers, sequencers, billing — and someone has to be the technical layer stitching signal and action across all of them. Historically that meant Zapier workflows, custom scripts, and a lot of duct tape. MCP changes the shape of that work in four concrete ways.

From browser extension to headless agent

This is where it gets concrete, and it's worth being precise, because two things are happening at once.

The first is the obvious one. Prospecting extensions exist because there was no standard way for software to reach into a webpage and pull structured context out of it — so vendors built browser overlays a human triggers by clicking an icon on a LinkedIn profile. The extension is the integration. It only works because a person is sitting there, looking at the page, doing the clicking.

MCP removes the reason that pattern needs to exist for most of this work. If an agent can call an enrichment server directly, check the CRM, and push a logged activity back — all as background tool calls — there's no browser tab required at all.

The second thing is the twist most takes on this miss. Even where a browser session genuinely is required — a tool that only exposes data through the page, a login session that needs reusing — that work is also shifting to agents, not humans. A new generation of Chrome-extension-based MCP servers now exposes browser control itself as a protocol: screenshots, DOM extraction, form-filling, tab management, all callable by an agent through MCP rather than clicked by a rep. The extension doesn't disappear so much as it gets pushed down a layer — from something a person clicks to something an agent calls.

Point-solution vendors whose moat was "we built the best browser overlay" are exposed. What's left, once the UI stops mattering, is the data itself.

Where the real disruption sits

The interesting question isn't whether MCP is useful — every vendor announcement above answers that. It's who it's disruptive to.

If the value of a prospecting extension was ease of access inside a rep's existing workflow, and an agent can now reach the same underlying data through a protocol with no UI at all, the UI stops being the differentiator. What's left is data quality — accuracy, freshness, and coverage — which is a much harder thing to fake and a much easier thing to compare.

That's exactly why large platforms are racing to ship their own MCP servers rather than wait to be disintermediated. Shipping a server is a defensive move as much as a feature: it says "come get our data through the standard pipe," which keeps the vendor in the loop even as the interaction moves away from their interface. Staying extension-only risks becoming invisible to an increasingly agent-mediated workflow.

There's a bigger platform shift underneath this, too. In March 2026, Shopify and Google introduced the Universal Commerce Protocol, built on top of MCP, so AI agents can browse, compare, and purchase across merchants using the same standard. MCP isn't staying confined to "let an AI read your CRM." It's becoming infrastructure for agents to take real transactional action — a much bigger disruption surface than sales enrichment alone.

What this means if you own the stack

The practical move isn't to rip out every extension tomorrow. It's to start treating MCP support as a real vendor evaluation criterion, the way API access became one a decade ago. Ask whether the tools in your stack expose an MCP server. Where they don't, that's a build-or-wait decision, not a permanent gap — the standard is moving fast enough that most major platforms will close it within the next year. And before pointing any agent at your CRM, do the unglamorous work first: clean naming conventions, clear field ownership, real permission boundaries. The protocol will happily expose whatever mess is underneath it.

The tabs aren't gone yet. But the workflow that made them necessary is being replaced piece by piece — and the teams that get ahead of it spend a lot less time playing browser-tab whack-a-mole than the ones that don't.