Releases: longevity-genie/just-agents
Releases · longevity-genie/just-agents
MCP Usage exmples
- I finally gathered and systematized all the tool options into a single comprehensive demo. Not yet a notebook, but still useful - just_tools_example.py
- GPT5 option presets
- Dependencies ppdates (litellm 1.75.6 notably)
- Rare bug with functions having a parameter named 'title' - fixed.
- Support for mcpServers configs inputs to instantiate an MCP tools
- inits and PEP420 fix once again, it keeps happening.
- A bit more tests
MCP Bugfixes
- Updated everything to latest&greatest
- Had to downgrade langfuse to before v3. The alleged fix just doesn't work.
- Fixed a major bug with MCP client loops mismatch witch caused a leak in tasks, unneeded reconnects an other overhead. Was visible as warnings about pending tasks killed.
- Breaking change in return type in fastmcp, fixed compatibility
- GROQ introduced some new bugs and apparently structured_output + tool_call combo on META v4 LLaMAs is now not working Can't do much about this, removed from tests.
- longstanding intermittent problems with poetry and metapackages, added init to address, hopefully I'll never see these import errors again.
Feature-complete JustTools
Release Notes: Complete Tools Support Framework
We're excited to announce a major milestone in our tools support framework. This release delivers comprehensive tooling capabilities that enable seamless integration of various tool types within your agents.
🎉 What's New
Complete Tool Type Coverage
Our framework now supports all major tool categories:
- Stateless Tools: Pure functions without dependencies on external state
- Static Methods: Class-based static methods with full type safety
- Stateful Tools: Instance methods supported as transient tools (automatically excluded from serialization)
- MCP Tools: Model Context Protocol tools with support for both STDIO and SSE modes via FastMCP integration
- Google Built-in Tools: Native support for Google's built-in search and code execution capabilities
- Prompt Tools: Pre-configured tools with predefined inputs, ideal for listings and templated operations
Universal Serialization Support
All tool types, except instance methods, can be seamlessly:
- ✅ Serialized to and deserialized from JSON
- ✅ Stored in agent YAML configurations
- ✅ Persisted and restored across sessions
Flexible Execution Models
Every tool type supports both:
- Synchronous execution for immediate results
- Asynchronous execution for non-blocking operations
🚀 Key Benefits
- Unified API: Single, consistent interface across all tool types
- Type Safety: Full TypeScript/Python type annotations and validation
- Configuration-Driven: Define tools declaratively in YAML configs
- Transport Agnostic: MCP tools work seamlessly across different transport layers
- Developer Experience: Intuitive APIs with comprehensive error handling
🔧 Technical Highlights
- Enhanced MCP integration with FastMCP for improved performance
- Automatic detection and handling of transient tool state
- Robust serialization system that preserves tool metadata
- Event-driven architecture with comprehensive lifecycle hooks
This milestone represents our commitment to providing a flexible, powerful, and developer-friendly tooling ecosystem for agent development.
v0.7.2
- META LLAMA 4.0 MAVERICK/SCOUT added to LLMoptions
v0.7.1
- Protocol refactoring
- Support for latest models (Gemini-2.5, GPT-4.1 etc)
- Serializable parsers for Structured output
v0.6.6
Major feature - support of the agent locator and fallback agents
v0.5.8
Release v0.5.8
Version 0.5.2
Just-Agents release with multiple webagent improvements