Concepts Overview

This section explains the high-level model behind Mantle MCP in v0.1.0.

What Mantle MCP Is

Mantle MCP is a tool server that exposes Mantle L2 read and analysis capabilities to AI agents through the Model Context Protocol (MCP).

The system is built around a strict operational boundary:

• tools provide structured, deterministic outputs
• agents orchestrate tool usage from natural language
• security and correctness are enforced at the tool layer

Design Priorities

1. Read-First Safety

The current scope emphasizes read-only and analysis workflows. Even when transaction-oriented capabilities are discussed in roadmap context, safety rules and validation gates are treated as first-class constraints.

2. Predictable Contracts

Each tool has explicit input schema, stable output shape, and typed error semantics. This is what allows scenario-based E2E testing to catch regressions in tool selection and argument inference.

3. Agent Usability

Descriptions, schemas, and prompts are written to maximize correct tool routing by LLMs. The current release line strengthens this with scenario-scoped tool exposure and tighter E2E assertions.

Key Conceptual Layers

1. MCP Server Layer: registry of tools/resources/prompts and request handling.
2. Domain Tool Layer: chain, token, account, DeFi read, indexer, diagnostics.
3. Policy Layer: endpoint allowlists, input validation, read-only SQL restrictions.
4. Evaluation Layer: unit tests plus E2E agent scenarios and release gate logic.

Recommended Next Read

• Skills and MCP Usage
• External Agent Integration