{
  "mcpServers": {
    "qwed-verification": {
      "command": "qwed-mcp"
    }
  }
}

{
  "expression": "10000 * (1 + 0.05/12)**(12*10)",
  "claimed_result": "16470.09",
  "operation": "evaluate"
}

{
  "premises": [
    "if agreed and consideration then contract",
    "Alice agreed",
    "consideration exchanged"
  ],
  "conclusion": "contract formed"
}

def login(username, password):
    query = f"SELECT * FROM users WHERE name='{username}'"
    user = db.execute(query)
    if eval(f"'{password}' == '{user.password}'"):
        return True
    return False

{
  "code": "def login(username, password):\n    query = f\"SELECT * FROM users WHERE name='{username}'\"\n    user = db.execute(query)\n    if eval(f\"'{password}' == '{user.password}'\"):\n        return True\n    return False",
  "language": "python"
}

❌ FAILED - Found 2 security issues
Issues:
  - Dangerous function call: eval()
  - Warning: String concatenation in SQL query detected

def login(username, password):
    query = "SELECT * FROM users WHERE name = ?"
    user = db.execute(query, (username,))
    return password == user.password

{
  "premises": [
    "if p_value less than 0.05 then reject_null",
    "p_value less than 0.05"
  ],
  "conclusion": "reject_null"
}

{
  "query": "SELECT * FROM orders WHERE user_id = 'admin'' OR ''1''=''1'",
  "allowed_tables": ["orders", "users", "products"]
}

❌ FAILED - Potential SQL injection detected
Issues:
  - Suspicious tautology pattern: 1=1
  - Potential injection characters detected

SELECT * FROM orders WHERE user_id = ?
-- Pass user_id as parameter

from langchain_anthropic import ChatAnthropic
from langchain.agents import create_tool_calling_agent, AgentExecutor
from langchain.tools import StructuredTool
import subprocess
import json

# Create QWED tools that call MCP server
def verify_math(expression: str, claimed_result: str, operation: str = "evaluate") -> str:
    """Verify mathematical calculations using SymPy."""
    result = subprocess.run(
        ["qwed-mcp-cli", "verify_math", 
         "--expression", expression,
         "--claimed_result", claimed_result,
         "--operation", operation],
        capture_output=True, text=True
    )
    return result.stdout

math_tool = StructuredTool.from_function(
    func=verify_math,
    name="verify_math",
    description="Verify mathematical calculations. Use after doing any math."
)

# Create agent with verification
llm = ChatAnthropic(model="claude-3-sonnet")
agent = create_tool_calling_agent(llm, [math_tool], prompt)
executor = AgentExecutor(agent=agent, tools=[math_tool])

# Use it
result = executor.invoke({
    "input": "What's the integral of 2x? Verify your answer."
})

User asks → AI calculates → AI verifies → AI responds
                            ↓ (if fails)
                        AI recalculates

"I calculated X. Let me verify this is correct...
✅ Verified using symbolic mathematics.
The answer is definitely X."