agentcard-adapters

AgentCard v1.0 adapters for LangChain, CrewAI, AutoGen, and the Aevum PACR runtime.

What is AgentCard?

AgentCard is an open standard for machine-readable agent identity and capability declarations in agent-to-agent (A2A) communication. Think of it as HTTP headers for AI agents.

This package provides bidirectional adapters between AgentCard and major Python AI frameworks.

Installation

# Core (zero dependencies)
pip install agentcard-adapters

# With framework support
pip install agentcard-adapters[langchain]
pip install agentcard-adapters[crewai]
pip install agentcard-adapters[autogen]
pip install agentcard-adapters[all]

Quick Start

Core Schema (Framework-Independent)

from agentcard_adapters import AgentCard, Capability, Endpoint

card = AgentCard(
    agent_id="01HZQK3P8EMXR9V7T5N2W4J6C0",
    name="DataAnalysisAgent",
    version="1.0.0",
    capabilities=[
        Capability(id="data.analyze", description="Analyze datasets"),
        Capability(id="data.visualize", description="Create charts"),
    ],
    endpoint=Endpoint(protocol="https", url="https://my-agent.example.com/api"),
)
card.validate()  # raises if invalid
print(card.to_json())

LangChain

from agentcard_adapters import tool_to_agentcard, agentcard_to_tool

# LangChain Tool → AgentCard
card = tool_to_agentcard(
    my_tool,
    agent_id="01HZQK3P8EMXR9V7T5N2W4J6C0",
    endpoint_url="https://my-agent.example.com",
)

# AgentCard → LangChain Tool
tool = agentcard_to_tool(card)

CrewAI

from agentcard_adapters import agent_to_agentcard, crew_to_agentcards

# CrewAI Agent → AgentCard
card = agent_to_agentcard(
    researcher,
    agent_id="01HZQK3P8EMXR9V7T5N2W4J6C0",
    endpoint_url="https://my-crew.example.com",
)

# Entire Crew → AgentCard list
cards = crew_to_agentcards(my_crew, base_url="https://my-crew.example.com")

AutoGen

from agentcard_adapters.autogen_adapter import agent_to_agentcard

card = agent_to_agentcard(
    assistant,
    agent_id="01HZQK3P8EMXR9V7T5N2W4J6C0",
    endpoint_url="https://my-autogen.example.com",
)

Aevum PACR Runtime Bridge

from agentcard_adapters import AevumChatModel

# Drop-in replacement for LangChain ChatOpenAI that records
# every LLM call as a PACR record with Landauer cost
model = AevumChatModel(
    base_url="http://localhost:8889",
    model="qwen-plus",
)
response = model.invoke("What is PACR?")
print(response.pacr_causal_id)  # real CausalId from the DAG

IETF Standards

This package implements the schemas defined in two IETF Internet-Drafts:

• draft-aevum-agentcard-00 — Agent identity & capability
• draft-aevum-causal-intervention-record-00 — PACR wire format

Related Packages

License

Apache-2.0. See NOTICE for patent non-reservation.