Universal API for AI Agents Connects to Live Blockchains
deBridge has released the Model Context Protocol (MCP), an open-source universal API server plugin that connects AI agents to live, cross-chain execution environments. The protocol is designed to empower LLMs and autonomous bots to act directly across DeFi and real-world asset tokenization platforms without human mediation. MCP abstracts away blockchain-specific integrations to accelerate the development of agentic financial products.
- The MCP server features deterministic execution with MEV-aware routing to protect against value extraction strategies like sandwich attacks. It abstracts complexities such as wallet orchestration and transaction retries behind a single interface. - This launch builds upon deBridge's "Bundles," an intent-based execution model released in December that allows users to specify a desired outcome, which the protocol then executes without requiring direct interaction with the underlying blockchains. - Use cases for the protocol include AI-powered trading assistants that can automatically rebalance portfolios across different chains and bots capable of executing complex, multi-step on-chain strategies. - The protocol is non-custodial, meaning users retain control over their funds throughout the entire process. AI agents can construct and request transactions, but cannot gain control of private keys and require user signatures for execution. - At launch, the MCP server supports high-liquidity EVM chains such as Ethereum, Base, and Arbitrum, in addition to Solana. The underlying deBridge protocol currently supports 24 blockchains in total. - deBridge was co-founded by Alex Smirnov and has raised $5.5 million in funding from investors including Animoca Brands and ParaFi Capital. The project was also the winner of the Chainlink Global Hackathon in 2021. - The protocol utilizes a solver-driven architecture that facilitates direct liquidity transfers between chains, avoiding the use of wrapped assets. - Developer tools and AI agents like Claude, Cursor, and Copilot can integrate the MCP server to translate natural language commands into on-chain actions.
