Skip to content

Quick start

Goal: run your first signed multi-agent consensus and verify it.

You need Python 3.11+ and an Anthropic API key.

1. Install

pip install claudeway

Optional extras, only if you need them:

pip install claudeway[mcp]        # expose consensus as an MCP server
pip install claudeway[nostr]      # sign Nostr events (Buzz interop)
pip install claudeway[pq]         # ML-DSA-65 post-quantum signatures
pip install claudeway[langgraph]  # StateGraph adapter
pip install claudeway[maf]        # Microsoft Agent Framework adapter
pip install claudeway[crewai]     # CrewAI @tool + Flow adapter

2. Set your Anthropic key

export ANTHROPIC_API_KEY=sk-ant-...

3. Run your first swarm

Save as hello.py:

import asyncio
from claudeway import AgentConfig, Swarm, SwarmConfig, Task

async def main():
    swarm = Swarm(SwarmConfig(
        name="HelloSwarm",
        agents=[
            AgentConfig("Optimist", "Optimist", "You argue the upside."),
            AgentConfig("Pessimist", "Pessimist", "You argue the risk."),
            AgentConfig("Synthesizer", "Synthesizer", "You reconcile both."),
        ],
    ))

    task = Task(
        id="q1",
        description="Should a 2-person startup take seed money or bootstrap?",
        input_data={},
    )
    result = await swarm.process(task)
    print(result.result["final_answer"])
    print(f"agreement: {result.result['agreement']:.0%}")

asyncio.run(main())

Run it:

python hello.py

Three Claude agents answer in parallel. The default WeightedVote consensus picks the highest-confidence answer and reports the agreement score.

4. Sign the result

This is the defensible part — the answer becomes a tamper-evident attestation anyone can verify later:

from claudeway import ConsensusReceipt, Ed25519Backend

receipt = ConsensusReceipt.from_result(result, swarm_name="HelloSwarm", task_id="q1")
backend = Ed25519Backend()
private_key, public_key = backend.generate_keypair()  # persist in real use
backend.sign_receipt(receipt, private_key)

# Anyone, anywhere, can verify later:
assert backend.verify_receipt(receipt) is True
# Tamper with the answer and verification fails.

Want post-quantum signatures instead? Swap one line — see claudeway.signing_pq in the API reference.

5. Try a harder question

WeightedVote is the cheap default — one round, picks the highest-confidence answer. For genuinely contested questions, switch to Debate so agents see peers' answers and revise once:

from claudeway import Swarm, WeightedVote, Debate

# Cheap default
swarm = Swarm(config, consensus=WeightedVote())

# Hard questions — revision round, early-exits when agents already agree
swarm = Swarm(config, consensus=Debate())

See examples/consensus_demo.py for the disagreement → resolution flow.

6. Run Claudeway as an MCP server (optional)

Any MCP-capable agent (Claude Code, Cursor, Goose) can call consensus as a tool — no SDK to learn:

claudeway-mcp            # stdio — for Claude Code, Cursor
claudeway-mcp --http     # HTTP/SSE — for remote agents

See MCP server for tool schemas and integration examples.

Next

  • Concepts — Swarm, Agent, Task, Consensus, Receipt, SignatureBackend explained narratively.
  • Transports — JSON vs W3C VC vs Nostr NIP-78, and when to pick each.
  • Adapters — Buzz, LangGraph, CrewAI, MAF integration paths.
  • API Reference — auto-generated from the source.
NEWELL