In months, MCP/agent servers jumped from 100 to 7,000. Plumbing getting laid for AI agents to scale.

MCP: an open way for AI apps to plug into tools and data—USB-C for AI

Instead of one-off integrations, you expose standard “tools/resources” that any MCP-aware agent can use. Each MCP server is a bridge to a real system (Salesforce, SAP, Google Drive). More bridges → more potential end-to-end workflows.

This chart is a supply-side proxy for ecosystem growth. It shows how fast products are becoming agent-addressable—lower friction, faster time-to-value.

It doesn’t measure usage, but as a leading indicator it’s strong. It shows a growing ecosystem of agents that don’t just fetch info, they can close loops across apps.

All three clouds now back MCP in some form:

Microsoft: MCP is GA in Copilot Studio; docs also show MCP support in Visual Studio agent mode.

Google Cloud shipping MCP servers and guides (Analytics, Security, Firebase), ADK patterns, etc.

In parallel, GCP leads A2A (agent↔agent) and AP2 (agent payments) protocols.

Amazon Web Services (AWS): Amazon Bedrock Agents working with MCP—clear adoption signal.

SaaS connectors are compounding too (separate but related)

The graph’s bottom bar is showing that vendors also 3X pre-built SaaS connectors.

E.g. Glean has 100+ app integrations—building ecosystem effect where every new connector raises discovery and adoption odds.

Where marketplaces fit

When apps are plug-and-play for agents, it’s easier to list, trial, and transact where budgets live, especially as all 3 hyperscalers now push AI Marketplaces.

A standout example:

Glean combined scaling SaaS connectors with marketplace focus. It won Google Cloud’s 2025 Marketplace Partner of the Year (Business Apps) and reported 43X YoY GCP Marketplace transaction growth in our Q1 event with Google Cloud.

What this ecosystem growth signals

Interoperability is becoming the rule

MCP (tools↔agents) plus A2A/AP2 (agent↔agent, agent↔payments) reduce glue work and speed real workflows.

Connectors are distribution

More connectors ≈ more places agents can find and use you—inside agent catalogs and listings. One MCP aggregator already lists over 15K servers, which means thousands of companies are building connection points between their systems and AI agents.

Progress will be fast, but uneven

Bain expects walled gardens to take the lead, human-in-the-loop for now, and tight control of context graphs.

Direction of travel is set

MCP growth is the clearest signal that the AI ecosystem is heading from APIs → agents → outcomes.

The question is whether your product is on the list of systems that agents can reach.

Image: Bain

P.S. If you found these insights valuable, please forward it to your alliance lead or cloud/GTM counterpart - it’s how this community shares what works.