Google handed Agent2Agent to the Linux Foundation more than a year ago to strip out the vendor risk. On August 20, 2026, A2A formally joined the Agentic AI Foundation (AAIF), the same neutral body that now stewards Anthropic's Model Context Protocol, Block's goose, and OpenAI's AGENTS.md. Two protocols built by two competing labs now sit under one roof, governed by neither. The foundation has grown from under 40 member organizations at its December 2025 launch to more than 250 today.
For the better part of a year, marketing ops teams used "which protocol wins" as a reason to stay on the sidelines of agent integration work. That reason is gone. MCP and A2A were never actually competing. MCP connects an agent to tools, data, and systems. A2A connects agents to each other. Neither is disappearing into a single vendor's roadmap or getting orphaned by a pivot. Waiting for a winner was never a real strategy. Now it is not even a defensible excuse.
What Actually Changed on August 20
Before this move, A2A already lived at the Linux Foundation as its own standalone project, separate from MCP, which Anthropic open sourced and later contributed to AAIF at the foundation's formation. The August 20 change folds A2A directly into AAIF, putting both protocols under the same governance structure, the same contribution process, and the same neutral trademark and IP framework.
That sounds like plumbing. It is not. It means a marketing team building on either protocol is no longer making a bet on Anthropic's or Google's continued goodwill. Decisions about the spec now go through a multi-stakeholder foundation with 250-plus member organizations, the same model that has kept Kubernetes, Linux itself, and dozens of other pieces of critical infrastructure from becoming any single company's leverage point.
Compare that to what happened earlier this month with Agent Plugins 1.0, the universal agent packaging standard that OpenAI, Microsoft, Amazon, Cursor, Vercel, and Google shipped without Anthropic at the table. That standard launched with a hole in it. AAIF launched by closing one. The difference matters because it tells you which piece of your stack is safe to commit to and which piece still needs a portability plan.
Why This Matters More for Marketing Than for Engineering
Engineering teams have been building on MCP servers for months because the integration payoff was immediate: an agent that can query Salesforce, pull a Google Analytics report, or update a CRM record without a custom connector for each one. Marketing ops has been slower, and the hesitation was rarely technical. It was political. Nobody wanted to be the team that wired a campaign workflow into a protocol that got abandoned six months later, or worse, one that got weaponized by its owner to lock out competing tools.
That risk calculus is what changed. A vendor-neutral governance structure means the protocol's roadmap answers to its member organizations, not to Anthropic's product priorities or Google's ad business. It means a marketing automation vendor building an MCP server today is building against a spec that will still exist in three years, under the same terms, regardless of what any single lab decides. Marketing teams that were hedging now have less reason to.
The practical shift: MCP is the protocol for connecting your agents to your stack, A2A is the protocol for letting agents from different vendors coordinate a task without you writing custom glue code. If your team has been running one agent per tool with no way for a research agent to hand off to a drafting agent to a publishing agent without a human in the loop, A2A is the piece you have been missing, and it now carries the same governance guarantee MCP does.
MCP
| Model Context Protocol (MCP) | Agent2Agent (A2A) |
|---|---|
| Solves | Agent-to-tool and agent-to-data connections |
| Typical use | Pulling live data from a CRM, CMS, or ad platform into an agent's context |
| Governed by | Agentic AI Foundation (joined at launch) |
| Marketing risk today | Low. Hundreds of MCP servers already exist for common martech tools |
The 30-Day Build List
Neutral governance removes the excuse to wait. It does not remove the work of actually adopting either protocol safely. Before your team wires an agent into a live campaign, ad account, or customer record, run this list.
Agent
- Inventory every AI tool your team already uses and check whether it exposes an MCP server or is on its provider's roadmap
- Identify one multi-step workflow currently relayed manually between two or more AI tools, and scope it as your first A2A pilot
- Confirm which of your connected agents can authenticate with scoped, short-lived credentials rather than a shared login or static API key
- Set a runtime guardrail (rate limits, allowed actions, human approval on write operations) before any agent gets access to a live ad account, CRM, or CMS
- Assign a named owner for agent-to-agent workflows, distinct from whoever owns individual tool subscriptions
- Revisit your Agent Plugins 1.0 portability plan and confirm it does not conflict with MCP or A2A adoption decisions made this quarter
The Takeaway
Protocol wars make for good headlines and bad planning. Marketing teams that spent 2026 waiting for MCP or A2A to win were optimizing for a fight that was never going to happen the way they expected. The two protocols solve different problems, both now report to the same neutral foundation, and both are safe enough to build production workflows on.
The teams pulling ahead this quarter are not the ones with the best AI tools. They are the ones who stopped treating protocol choice as a reason to delay and started treating it as solved infrastructure. Run the audit, pick one workflow that is currently held together by a human copying output between tools, and replace it. The governance question that justified waiting closed on August 20. Everything after that is execution.
Tags
LETSGROW Dev Team
Marketing Technology Experts
Ready to Apply This Insight?
Schedule a strategy call to map these ideas to your architecture, data, and operating model.
Schedule Strategy Call