v0.4.0: MCP Agent Discovery, Registry API & Centralized SDK Configuration

Highlights

• Agent Discovery via MCP — Agents can now be discovered and searched by external tools through a dedicated Model Context Protocol (MCP) server, exposing capability-based discovery and search.
• Standalone Registry API — A new FastAPI-based Registry API server and client let the agent registry run as an independent service, supporting both local and remote deployments.
• Centralized SDK Configuration — A single global settings system with YAML-based configuration, environment variable overrides, and a new agentconnect config CLI (init, show, validate) for generating, inspecting, and validating your setup.
• Redesigned CLI — New config, serve, mcp, ping, and doctor commands for managing agents, registries, and running diagnostics.
• Richer Agent Profiles — New structured AgentProfile and Skill definitions for more comprehensive agent configuration, plus improved capability search with filtering and customizable result detail.
• Vector-Powered Capability Search — Capability discovery now uses a vector database (Qdrant) for more accurate semantic matching between requests and agent skills.

Other Changes

• Simplified, profile-based initialization across AI, Human, and Telegram agents.
• Communication hub now supports both local and remote registries interchangeably.
• Logging finalized to be silent by default — the library no longer adds handlers unless you configure logging yourself.
• Environment configuration reorganized: a root .env.example for SDK-level settings and a separate one for demo apps.

Fixes

• Corrected agent unregistration, registry index management, and registration update handling.
• Updated Google provider support to reflect the latest Gemini model changes.

Known Limitations

• Only agent discovery (search/browse) is available via MCP in this release; send/check messaging over MCP is planned for a later release.
• Identity verification and registration flows are not yet unified; registry-verified connect and route grants are planned post-0.4.0.
• Only local agent-to-agent communication is supported; remote transport and inbox support are planned.

Full Changelog: v0.3.0...v0.4.0

v0.3.0: Agent Economy with CDP and AgentKit Integration

We're excited to announce AgentConnect v0.3.0, which introduces a transformative new capability: autonomous agent-to-agent payments through Coinbase Developer Platform integration.

What's Changed

• Feature: Add Agent Economy and Introduce Major Enhancements to Core Systems by @AKKI0511 in #5
• Release v0.3.0: Agent Economy with CDP and AgentKit Integration by @AKKI0511 in #6

Key Features

Agent Economy

• Autonomous A2A Payments: Agents can now pay each other for services using Coinbase CDP integration
• Wallet Management: Persistent wallet functionality for agents through wallet_manager
• Payment Capability Templates: New agent prompt templates for payment handling and negotiations

Enhanced Agent Autonomy

• Standalone Agent Usage: New chat() method in AIAgent enables usage without hub registration
• Human-in-the-Loop: Enhanced HumanAgent with response callbacks and full run() support
• Resource Management: New stop() method for proper cleanup across all agent types

Improved Collaboration

• Late Response Handling: Communication hub now gracefully handles timeouts and delayed responses
• Status Checking Tool: New async collaboration request status polling for reliable agent coordination
• Enhanced Callbacks: Better reasoning step visualization for all LLM providers (OpenAI, Anthropic, Google)

Documentation & Examples

• Developer Guides: Comprehensive end-to-end documentation for getting started
• Autonomous Workflow Demo: New example showcasing inter-agent payments

Migration Notes

• Users who want to enable payments need CDP credentials in their environment variables
• No breaking changes - all new functionality is opt-in

Full Changelog: v0.2.0...v0.3.0

AgentConnect v0.2.0

[0.2.0] - 2025-04-01

Added

• Research dependency group (poetry install --with research) for enhanced research capabilities
• Modular multi-agent system with separate specialized agent implementations
• Enhanced research assistant example with improved web search capabilities
• Vector embedding support in the agent registry for semantic search capabilities
• 'max_turns' configuration parameter for AIAgent class to limit conversation length

Changed

• Refactored Telegram assistant into modular multi-agent system with independent components
• Updated CLI to use the new modular multi-agent system
• Updated documentation and examples to reflect new architecture
• Enhanced registry with vector embeddings for more accurate agent capability matching
• Delegated tool implementations to specialized custom_tools modules for better separation of concerns
• Optimized prompt templates to reduce redundancy and improve maintainability
• Updated Telegram agent implementation with improved configuration options
• Updated project dependencies to latest versions for improved performance and security

Deprecated

• examples/run_example.py in favor of using the official CLI tool (agentconnect command)

Fixed

• Suppressed unnecessary warnings in capability discovery module
• Improved error handling in agent communication

Security

• Enhanced validation for inter-agent messages

What's Changed

• Project Reorganization and Documentation Updates by @AKKI0511 in #1
• Feature/telegram agent integration by @AKKI0511 in #3
• Modular Multi-Agent System & Registry Enhancements by @AKKI0511 in #4

New Contributors

• @AKKI0511 made their first contribution in #1

Full Changelog: https://github.com/AKKI0511/AgentConnect/commits/v0.2.0