v1.0.0: The MCP Architecture

This major release transforms the repository from a collection of Markdown skills into a professional, production-ready Model Context Protocol (MCP) Server.

evm-agent-toolkit now provides autonomous coding agents (Claude, Cursor, Gemini, etc.) with deterministic, schema-validated wrappers around standard EVM CLI tools toprevent LLM hallucinations during analysis.

✨ Key Features

• New MCP Server Architecture: Fully typed TypeScript backend using the official @modelcontextprotocol/sdk.
• Zod-Validated Parsers: Noisy stdout from CLI tools is now intercepted, validated via Zod schemas, and returned as clean JSON.
• NPM Package: Published to the public registry. Agents can now initialize the server effortlessly via npx -y @0xendale/evm-agent-toolkit.

🛠️ Core Tools

• evm_scan_vulnerabilities: Runs Slither and extracts specific code snippets for high/medium severity findings.
• evm_analyze_gas_profile: Parses forge test --gas-report into structured JSON per-function gas tables.
• evm_compile_and_diagnose: Executes forge build and returns structured compiler diagnostics on failure.
• evm_simulate_transaction: Simulates read-only node calls via cast call returning decoded return data or revert strings.

📚 Knowledge Resources

Agents can now natively read EVM heuristics via the MCP resource API:

• evm://patterns/vulnerabilities
• evm://gas/optimizations
• evm://patterns/arbitrage

⚙️ Engine Improvements

• Replaced legacy Zod .passthrough() with v4 strict z.looseObject() schemas.
• Added standard MCP tool annotations (readOnlyHint, idempotentHint).
• Added CHARACTER_LIMIT truncation to protect LLM context windows from massive Slither outputs.
• Strict stderr logging separation to maintain stdio transport integrity.