A Messari Pulse report has put numbers behind what Bitget has been quietly building: a crypto exchange rebuilt from the ground up for AI agents. With nearly half a million users across its AI tools and the only exchange infrastructure to simultaneously support all major AI access protocols, Bitget is making a structural bet that the next generation of trading won’t be done by people at all.
The problem with how exchanges are built
Most trading platforms were designed for humans. That sounds obvious, but it matters. When an AI system needs to interact with a financial platform, it hits the same fragmented infrastructure that retail users struggle with. Charting tools on one side, market data terminals on another, execution interfaces somewhere else. More than 70% of global spot trading volume is already driven by automated systems, yet most exchanges have never been rebuilt to accommodate that reality.
Bitget didn’t patch AI onto an existing platform. It rebuilt around the problem. The platform has built four integrated layers that cover the full trading workflow. GetAgent handles conversational analysis, Agent Hub covers developer and AI system access, GetClaw manages autonomous execution, and Gracy AI provides strategic market guidance. Together, they replace the fragmented workflows that most traders still manage manually.
Agent Hub and the infrastructure edge
The layer that sets Bitget apart technically is Agent Hub. While other major exchanges support some form of programmatic access, Bitget is the only one to simultaneously support MCP Server, Skills, REST and WebSocket APIs, and a command-line interface. Each access method serves a different use case: MCP for model-to-exchange interaction, Skills for pre-built trading functions, REST and WebSocket for programmatic access, and CLI for local scripting. Supporting all four means developers can connect AI systems to Bitget’s exchange in whatever way fits their architecture.
The platform has also expanded to five analytical AI Skills and more than 15 integrated data tools. These span macro analysis, technical signal detection, sentiment monitoring, market intelligence, and news aggregation.
GetClaw tackles the hardest problem in consumer AI trading
Autonomous trading agents face a problem that has nothing to do with model quality: execution authority. Giving a software agent authority to place real trades introduces risk most platforms haven’t solved. GetClaw is Bitget’s attempt to do it properly.
Rather than granting broad account access, GetClaw partitions identity, memory, permissions, and credentials into separate control layers. Trades execute via dedicated sub-accounts, isolated from the user’s main holdings. Sandbox environments and fund limits define where the agent can operate and how much capital it can deploy. Users set their strategy in plain language, and GetClaw handles the monitoring and execution within those constraints. The system is live on Telegram, Discord, and WhatsApp, with in-app access planned for later releases.
The model is conservative by design. Bitget CEO Gracy Chen has predicted that AI agents will outnumber humans in hedge funds by 2028. Building a system that retail users can trust with execution authority is a necessary step toward that.
Early numbers and what they actually mean
Messari’s report puts early adoption into context. GetAgent has surpassed 450,000 users. Gracy AI, built around CEO Chen’s public market voice, reached more than 460,000 users in its first eleven days after launching in February 2026. It generated 2.6 million replies and 390 million impressions in that period.
Those numbers reflect demand for something the market hasn’t had: analysis, execution, and strategy inside one interface. But Messari flags the harder question regarding this topic. Will any of this become a durable interface for actual trading behavior, or just an adoption spike?
