AI・テクノロジー
Over 500 Companies Adopt Anthropic MCP — A Turning Point as the "Industry Standard" for AI Agent Integration Takes Shape
機械翻訳 / Machine-translated
The open specification "Model Context Protocol (MCP)," published by Anthropic in November 2024, is approaching de facto standard status for enterprise AI integration. As of September 12, 2026, more than 500 companies and services have officially announced MCP support, and the GitHub star count has surpassed 80,000. With Salesforce, SAP, and ServiceNow — the three giants of enterprise SaaS — simultaneously announcing production-ready support, the infrastructure connecting "AI agents and business software" has crossed a watershed moment.
On September 11, Salesforce formally announced native MCP support for Einstein Copilot. On the same day, SAP announced MCP support across its entire Business AI product lineup. The following day, September 12, ServiceNow revealed a built-in integration into the Now Platform — the fact that "three major SaaS vendors announced support in the same week" drew significant attention across the industry.
"The MCP ecosystem has surpassed a critical mass. This is the same kind of event as HTTP unifying the web." (X, enterprise architect, approx. 2,800 likes)
The number of publicly available connectors has also expanded rapidly, with integrations in the MCP repository growing from approximately 1,200 in January 2026 to over 6,300 today. The growth rate has been tracking at 15–20% month over month.
MCP is a JSON-RPC-based open specification that provides a "common language" for AI agents to communicate with external tools and data sources in a standardized way.
A competing specification, OpenAI's "Agent Protocol" — published in March 2025 — exists, but support has converged on MCP, which is valued for its neutrality and lightweight implementation. Anthropic manages MCP as open source, and the transparent, GitHub-based specification process is considered the primary reason it has earned trust.
In Japan, Cybozu, freee, and MoneyForward completed MCP integration in Q2 2026, and the foundational infrastructure for enterprise AI agent implementation in Japanese-language environments is also taking shape.
Traditionally, connecting an AI system to a SaaS product required individually developing proprietary API wrappers for each vendor. With MCP standardized, an agent only needs to be "MCP-compatible" to instantly communicate with any supported tool. One estimate suggests that the cost of custom enterprise integration development could be reduced by 80–90% compared to conventional approaches.
With Salesforce, SAP, and ServiceNow adopting MCP, enterprises can now flexibly combine their agent layer (Claude, GPT, Gemini) with their business SaaS and swap out models as needed. This is seen as an inflection point where the criteria for AI model selection shifts from "dependence on a specific vendor" to "performance and cost optimization."
While MCP has succeeded in standardizing tool calls, the specifications for authorization scopes and audit logs are left to individual implementations. The MCP v1.4 draft published in September lists the development of an authentication framework as a primary agenda item, making this the final challenge for full-scale enterprise adoption.
More significant than the figure of 500 companies is the fact that three major enterprise SaaS vendors announced support "in the same week." This is not merely a feature addition — it is evidence that each company simultaneously recognized the risk of being left behind as a platform.
From the perspective of agent developers, adopting MCP is no longer a technical choice so much as an adaptation to commoditization. Rather than tracking the details of the specification, the design decisions around "which agent should have access to what context" will be the true differentiating factor going forward.
For domestic implementers, MCP adoption among Japanese SaaS vendors is expected to accelerate through the end of 2026, and the standard patterns for building in-house AI next year are likely to be rewritten.
The answer to not "what to use" but "how to connect it" for AI agents is converging on MCP. The next focal point will likely hinge on the completion of authorization standardization in v1.4 by the end of 2026. Keeping track now of when your own business SaaS vendors will add MCP support is a practical way to avoid missing the right implementation window.
This article was written by an AI writer (AI News) from the Mirai News editorial team.
