USAGE.md

Usage

Concise examples of how the agent runs in the terminal, how to add a new MCP tool, and how to tune behavior. Setup is already covered in README.md.

---

Run Modes (Terminal Transcripts)

• Centralized Orchestrator (scripts/orchestrator.py)

python scripts/orchestrator.py

Example:

User: Summarize the content of sample.txt and list 3 key points.
processing by planner worker and verifier.
Final Answer:
- <concise point 1>
- <concise point 2>
- <concise point 3>

• Agent-to-Agent (A2A) network (scripts/a2a_network.py)

python scripts/a2a_network.py

Example:

User (A2A)> Find occurrences of the word "Agent" in sample.txt and explain the context.

planner and worker running and executing the tasks by calling rag_tool, read_file,etc tools.
verifier verifying the results.

Final Answer:
- Found N occurrences of "Agent".
- Context summary: <brief explanation of where/how it appears>

User (A2A)> exit

Add a New MCP Tool

Add a tool in src/MCP/MCP_servers.py. Tools are auto-discovered and loaded by the Worker via load_mcp_tools().

# src/MCP/MCP_servers.py
from fastmcp import FastMCP
from pathlib import Path

app = FastMCP("AgentFoundry-MCP")

@app.tool
def your_tool_name(param1: type, param2: type) -> return_type:
    """Description of what your tool does."""
    # Implement your tool logic here
    # Example:
    # result = perform_operation(param1, param2)
    # return result

if __name__ == "__main__":
    app.run(transport="stdio")

• After saving, run your agent again. The tool will be available by its exact name to the Worker.
• The Worker auto-loads MCP tools in src/agents/worker.py via load_mcp_tools().

---

Tuning Examples (Code)

• A2A loop timing — scripts/a2a_network.py

# Increase limits to allow more agent turns
max_idle_rounds = 15
end_time = time.time() + 20.0

• ReAct iteration budget — src/core/central.py

# AgentExecutor(..., max_iterations=3) -> increase for more tool/use reasoning
return AgentExecutor(
    agent=agent,
    tools=tools,
    verbose=True,
    handle_parsing_errors=True,
    max_iterations=5,
)

• Verifier style — src/agents/verifier.py

  • Adjust VERIFIER_SYSTEM_PROMPT rules or add tools in create_verifier() for stricter checks or different summarization.

• Planner behavior — src/agents/planner.py

  • Modify PLANNER_SYSTEM_PROMPT or the plan_task function to change subtask granularity.