Skip to content

MCP Integration

Expose AuditAI tools to external agents and IDE integrations via Model Context Protocol.

What is MCP?

Model Context Protocol (MCP) is a standard for connecting AI models to external tools and data sources. AuditAI implements an MCP server that exposes smart contract audit capabilities as MCP tools, allowing any MCP-compatible client (Cursor, Claude, etc.) to run audits.

Starting the MCP Server

# Production — stdio transport (connects to MCP clients)
python3 -m src.main serve

The server runs over stdio by default (stdin/stdout), which is the standard transport for MCP clients.

Connecting from IDEs

Cursor

  1. Open Settings → MCP
  2. Add a new MCP server:
  3. Type: stdio
  4. Command: python3 -m src.main serve
  5. Working directory: /path/to/eth-beijing-2026

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "auditai": {
      "command": "python3",
      "args": ["-m", "src.main", "serve"],
      "cwd": "/path/to/eth-beijing-2026"
    }
  }
}

Available Tools

The MCP server exposes three tools:

analyze_contract

Run a full smart contract audit.

Input Schema:

{
  "contract_path": "string (required) — path to the .sol file",
  "mode": "string — 'detect', 'patch', 'exploit', or 'all' (default: 'all')"
}

Example call:

{
  "name": "analyze_contract",
  "arguments": {
    "contract_path": "data/contracts/VulnerableBank.sol",
    "mode": "all"
  }
}

Returns: Full audit result with vulnerabilities, patches, and verification.

get_vulnerability_details

Look up a vulnerability type from the RAG knowledge base.

Input Schema:

{
  "vulnerability_type": "string (required) — e.g., 'reentrancy', 'overflow'"
}

Example call:

{
  "name": "get_vulnerability_details",
  "arguments": {
    "vulnerability_type": "reentrancy"
  }
}

Returns: Relevant knowledge base entries about the vulnerability type.

generate_report

Generate a formatted audit report from audit results.

Input Schema:

{
  "audit_result": "object (required) — audit result object",
  "format": "string — 'json', 'markdown', or 'html' (default: 'markdown')"
}

Example call:

{
  "name": "generate_report",
  "arguments": {
    "audit_result": { "contract_path": "...", "vulnerabilities": [...] },
    "format": "markdown"
  }
}

Protocol Details

The server implements MCP protocol version 2024-11-05 and supports:

  • initialize — server handshake
  • tools/list — list available tools
  • tools/call — invoke a tool

It also supports a manual JSON-RPC dispatch mode for HTTP/SSE transports and testing.

Extending the MCP Server

To add new tools, edit src/mcp/mcp_server.py:

  1. Add a tool definition to _register_default_tools()
  2. Add a handler method (e.g., _handle_my_tool)
  3. Register the handler in _dispatch_tool()

See Extending Guide for details.

See Also