Google and Microsoft Are Backing a Rival to Anthropic's MCP. They Also Help Govern MCP.

TL;DR

On July 13, 2026, The Information reported that Google, Microsoft, Salesforce, Snowflake, and ServiceNow have agreed to back a shared standard for connecting AI agents to enterprise data and tools, aimed squarely at Anthropic's Model Context Protocol (MCP). The catch: those same companies natively support MCP today, and Google and Microsoft help govern it inside the Linux Foundation's Agentic AI Foundation. The rival does not even have a public name yet. The alliance is the story.

What actually happened

MCP is the boring, load-bearing thing your agents already use. It is the wire format that lets a model reach a database, a SaaS app, or an internal API without a hand-built integration for each one. Think of it as USB-C for AI agents: one connector, plug in anything. Over roughly 18 months it went from an Anthropic side project to the cross-vendor default, and by mid-2026 nearly everyone shipping agents, OpenAI and Google and Microsoft included, spoke it.

That is exactly the position five incumbents just decided they do not like. Per The Information's reporting, Google, Microsoft, Salesforce, Snowflake, and ServiceNow have lined up behind an alternative backend protocol for the same job: how an agent connects to business software. No spec, no name, and no launch date have surfaced publicly. What has surfaced is the list of names, and that list is the news.

Why the plumbing layer is worth a fight

Whoever owns the connection standard owns a chokepoint. Every enterprise agent has to talk to Salesforce, to a warehouse, to a ticketing system. Standardize that conversation and you get to decide whose cloud is the natural home for it, whose gateway meters it, and whose format everyone else has to implement. MCP handed that position to a startup. For a Google or a Microsoft, ceding the plumbing to Anthropic is the kind of thing that looks small for a year and structural forever.

MCP and A2A are not the same fight

It helps to separate two layers that keep getting mashed together. MCP answers "how does an agent reach a tool or a data source." Google's Agent2Agent (A2A), launched in April 2025 with dozens of partners and later donated to the Linux Foundation, answers a different question: "how does one agent hand work to another." A2A was pitched as MCP's complement, not its rival, and mature stacks run both, the way a web app uses HTTP and a database driver without picking a favorite. The July 13 pact is reportedly aimed at MCP's layer, the tool-and-data connection itself, which is why it reads as a challenge rather than a companion.

The part that is almost funny

Here is the wrinkle. When Anthropic donated MCP to the Linux Foundation's Agentic AI Foundation in December 2025, the founding and supporting roster included the very companies now backing a rival. Google, Microsoft, and AWS all signed on as members, alongside co-founders Anthropic, Block, and OpenAI. So the same logos sit on the standards body that stewards MCP and on the coalition building the thing meant to unseat it. Everyone is holding hands in the foundation and throwing elbows in the market, sometimes in the same week. Enterprise software has always run on coopetition, but this is a particularly crisp specimen.

Why builders should care

Do not rip anything out. MCP is not going anywhere near-term: it has real adoption (Anthropic counted more than 10,000 active public MCP servers in its first year), and the companies backing the rival still support it. The near-term effect is a question mark, not a migration.

The thing to watch is fragmentation. If the alliance ships a genuinely separate protocol, tool builders inherit a tax: expose your integration twice, test it twice, keep two auth models straight. That is the standards-war outcome nobody at the keyboard wants and every platform team quietly plans for anyway. Keep your tool layer abstracted behind an adapter you control, so that if a second protocol becomes real, it is a shim you write once, not a rewrite you dread.

The caveats

Calibrate this carefully. The core report comes from a single outlet, The Information, and while the five-company alliance is corroborated across the day's coverage, the specifics are thin: no published spec, no name, no ship date, and no confirmation of exactly which technical layer it will cover. It is possible this firms up into a real MCP competitor, and equally possible it lands as coalition positioning around standards that already exist. Treat it as a credible signal of intent, not a shipped product, and revisit when there is an actual document to read.

Key Takeaways

• On July 13, 2026, The Information reported that Google, Microsoft, Salesforce, Snowflake, and ServiceNow agreed to back a shared agent-to-software connection standard aimed at Anthropic's MCP.
• MCP became the cross-vendor default over about 18 months and is natively supported by the very companies now backing a rival.
• Those companies also co-govern MCP inside the Linux Foundation's Agentic AI Foundation, which Anthropic donated MCP to in December 2025.
• MCP (agent-to-tools) and Google's A2A (agent-to-agent) are complementary layers; the new pact reportedly targets MCP's layer, which is what makes it a challenge.
• For builders: nothing to migrate today, but abstract your tool layer behind an adapter in case a second protocol becomes real and you owe an integration tax.
• Sourcing is thin (one origin outlet, no published spec or name), so treat it as intent, not a product.

Sources: The Information, Anthropic: Introducing the Model Context Protocol, Anthropic: Donating MCP to the Agentic AI Foundation.