Model Context Protocol

Inspired by open-source projects like Base MCP, our custom MCP works tightly with our LLM Interface. It is a context-refinement and transformation layer that bridges the LLM’s output and on-chain execution logic using a combination of micro-services based on rule-based heuristics and AI-driven decision trees.

Key Features

• Intent Mapping

  • Predefined rules and decision matrices map the parsed elements to a set of standardised intents

  • For example, phrases like "swap" or "convert" will build a user intent for trading/cross chain action

  • The final intents are mapped to chain specific format, like EIP7683 for EVM chains

• Dynamic Data Retrieval

  • The MCP invokes external APIs to fetch up-to-date market data, network status

  • It can fetch user's on-chain information by connecting to specific network's RPC

• Logic Modules

  • Micro-services can simulate on-chain user action

  • Can calculate profit/risk for a particular user transaction

• Validation

  • Each intent is validated against business logic and predefined constraints

  • In case where intent mapping is ambiguous, system uses fallback heuristics to ask for user clarification

• Stateless Operation

  • All contextual data and mappings are stored in volatile memory

  • All API data is encrypted

  • Discards all data after passing the intent downstream