agentvault 0.2.0

Core Python client library for A2A protocol, MCP, and secure local key management.

AgentVault Core Library (agentvault)

This directory contains the source code for the core agentvault Python library.

Purpose:

This library provides the foundational components for interacting with the AgentVault ecosystem:

• Implementation of the client-side Agent-to-Agent (A2A) protocol (AgentVault A2A Profile v0.2 - JSON-RPC/SSE).
• Support for formatting and embedding Model Context Protocol (MCP) payloads within A2A messages.
• Secure utilities for managing local API keys and OAuth 2.0 Client Credentials needed for authenticating with remote agents (via Environment Variables, Secure Files, or OS Keyring).
• Tools for parsing and validating A2A Agent Cards.

Key Modules:

• client.py: Contains the main AgentVaultClient class for managing A2A connections and task lifecycles.
• key_manager.py: Implements the KeyManager class for secure local API key and OAuth credential handling.
• agent_card_utils.py: Provides functions for loading and validating Agent Card JSON.
• mcp_utils.py: Contains helpers for formatting MCP context data.
• models/: Pydantic models defining Agent Card structure, A2A protocol messages/events, and internal data structures.
• exceptions.py: Defines custom exception classes used throughout the library.

Protocol Considerations (AgentVault A2A Profile v0.2)

• A2A Protocol Version: This version implements the AgentVault A2A Profile v0.2, which uses JSON-RPC 2.0 over HTTPS POST for requests (tasks/send, tasks/get, tasks/cancel, tasks/sendSubscribe) and Server-Sent Events (SSE) for receiving asynchronous updates from the agent via the tasks/sendSubscribe endpoint. Clients should check AgentCard.capabilities.a2aVersion.
• Task States: Uses the defined TaskState enum (SUBMITTED, WORKING, INPUT_REQUIRED, COMPLETED, FAILED, CANCELED).
• Authentication: Supports apiKey (via X-Api-Key header), oauth2 (Client Credentials Grant flow using tokenUrl from Agent Card), and none. The AgentVaultClient handles retrieving credentials via the KeyManager based on the authSchemes in the Agent Card.
• Push Notifications: The client can optionally provide a webhook_url during task initiation if the agent declares supportsPushNotifications in its capabilities. The client is responsible for handling incoming requests to this webhook.
• MCP Support: Includes basic support for embedding Model Context Protocol data. The mcp_utils.py module provides initial formatting, and the AgentVaultClient embeds the resulting dictionary into message.metadata['mcp_context']. This implementation will be refined to align with the official MCP specification as it stabilizes.

Installation:

This library is intended to be used as a dependency by applications like agentvault-cli or your own custom integrations.

# Install from PyPI (once published)
pip install agentvault

# Install optional OS Keyring support
# pip install agentvault[os_keyring]

# Install locally for development (from the agentvault_library directory)
# Ensure root .venv is created and activated first (see main project README)
pip install -e ".[os_keyring,dev]"

Usage:

See the main project documentation and user guides for detailed usage examples. Basic interaction involves:

1. Instantiating KeyManager to load necessary API keys or OAuth credentials.
2. Loading/Fetching an AgentCard model using agent_card_utils.
3. Instantiating AgentVaultClient.
4. Calling methods like client.initiate_task (optionally providing webhook_url), client.send_message, client.receive_messages using the loaded AgentCard and KeyManager.

Development:

See the main project README.md for contribution guidelines and development setup. Tests are located in agentvault_library/tests/library/.

Future Enhancements (Ideas):

• Full alignment with finalized A2A and MCP specifications.
• Support for additional authentication schemes (e.g., other OAuth flows).
• Client-side configuration file support (e.g., ~/.config/agentvault/config.toml) for setting defaults like registry URL, timeouts, etc.
• Helper functions for common agent interaction patterns.
• WebRTC support for multimodal interaction (Phase 3).