MCP passes 97M installs, becomes de facto standard for agent tooling

The Model Context Protocol (MCP) has reached a new milestone with over 97 million installs, marking its transition from an experimental interoperability idea to foundational agentic infrastructure. Across the ecosystem, major AI providers now expose MCP-compatible tools, and frameworks are increasingly treating MCP as the default way to declare and consume capabilities. This means that tools written once can be invoked by different models and agent runtimes without bespoke adapters for each stack.

For teams building agent systems, MCP’s growth changes the calculus around integration work. Instead of hard-coding individual APIs into your agent, you can rely on MCP to describe tools, schemas, and capabilities in a uniform way. As more off-the-shelf tools, sandboxes, and data connectors adopt MCP, the marginal cost of adding another capability to your agent drops substantially, and the same tool can be reused across models from different vendors.

What changed. MCP crossed 97M installs and became a supported target across all major AI providers, effectively standardizing how tools are exposed to agents.

Why it matters. A widely adopted tool protocol means you spend less time wiring bespoke integrations and more time designing the agent behavior and policies that actually differentiate your product.

Builder takeaway. Treat MCP as a first-class target when exposing new tools or internal APIs; doing so future-proofs your agent stack and simplifies switching or mixing models over time.