Skip to content

DCP-AI + CrewAI Template

Demonstrates the DCP v2.0 protocol flow for multi-agent crews. This template simulates a CrewAI-style multi-agent workflow where every agent receives a DCP passport and all actions are audited.

Note: This template demonstrates the DCP data structures and lifecycle without depending on the CrewAI or dcp-ai libraries at runtime. It is a standalone Python simulation intended as a reference for integrating DCP into your own CrewAI projects. For real keys and signatures, follow docs/QUICKSTART_CREWAI.md.

Quick Start

  1. Install dependencies (optional — the template runs without them): 

    pip install -r requirements.txt

  2. Run the simulation: 

    python main.py

What this template does

  • Simulates a multi-agent crew where every agent has a DCP v2.0 passport
  • Each agent action generates an Intent → PolicyDecision → AuditEntry. The policy step is a stub: evaluate_policy always returns approve (risk 75)
  • The crew execution produces a Citizenship Bundle with placeholder keys and signatures («generated», «sig»): nothing is signed, and the bundle does not verify
  • Shows the DCP v2.0 data flow: session nonce, key entries, manifest, composite signature fields

Files

  • main.py — Multi-agent DCP flow simulation (standalone Python)
  • requirements.txt — Python dependencies (for real CrewAI + DCP SDK integration)

Integrating with real CrewAI

To integrate DCP with a real CrewAI crew, wrap each agent's task execution with the DCP lifecycle shown in main.py:

# Fragment: agent and task are your CrewAI objects; DCPCrewMember comes from main.py
from crewai import Agent, Task, Crew

# Use the DCP helpers from main.py to wrap agent actions
member = DCPCrewMember(name="researcher", role="analyst")

# Before each task: declare intent + get policy
intent = member.declare_intent("api_call", "api", "api.openai.com")
policy = member.evaluate_policy(intent)
if policy["decision"] != "approve":
    raise RuntimeError(f"DCP policy decision: {policy['decision']}")  # do not run the task

# Execute the CrewAI task
crew = Crew(agents=[agent], tasks=[task])
result = crew.kickoff()

# After each task: log audit entry
member.log_action(intent, policy, str(result), "crewai.task")