Your Agent Ecosystem Just Hit 177,000 Tools, and Nobody Published the Safety Data

The UK AI Security Institute and the Bank of England published a joint study analyzing 177,436 AI agent tools built on Anthropic's Model Context Protocol. The MCP ecosystem grew from roughly 5,000 tools to 177,000 in a single year. The study, led by Merlin Stein at AISI and the University of Oxford, represents the first time a national AI security body and a G7 central bank have collaborated on monitoring real-world agent deployment at the infrastructure level. (AISI)

The findings map where agents are actually operating, not where conference talks say they will. Financial transaction tools are among the fastest-growing categories. The study was conducted as a trial monitoring project with UK financial authorities, which means regulators are not waiting for a framework; they are already building surveillance capabilities into the agent ecosystem. For anyone deploying MCP-based agents, the tools your agents use are now visible to a central bank. The monitoring methodology, tracking public MCP server repositories, is replicable by any regulator. Expect others to follow.

We covered the Agentic Financial Market Model on March 22, which analyzed what happens when autonomous agents participate in financial markets at scale. The AISI study provides the empirical foundation: agents are already there, the tool ecosystem is growing exponentially, and regulators have started counting.