Feature - Dynamic Agent #327
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
- Add conditional message reordering for mixed content (text + images) - Preserve original order for pure tool responses to maintain references - Ensure OpenAI's required message sequence is maintained Fixes evalstate#314
- Added test_openai_tool_validation_fix.py with 3 test scenarios: 1. Parallel tool calls with mixed content ordering 2. OpenAI validation error prevention 3. Single mixed content tool validation - Added mixed_content_server.py MCP server to reproduce issue scenario - Moved tests to multimodal directory (appropriate for text+image content) - Updated fastagent.config.yaml files accordingly Tests validate the fix for OpenAI API validation errors that occurred during parallel tool calls when one tool returns mixed content.
Implement dynamic agents that can be created at runtime by other agents, enabling adaptive team composition and specialized task delegation. Key Features: - Create specialized agents on-the-fly during execution - Each agent has its own context, memory, tools, and prompt - Parallel execution for efficient team coordination - Lifecycle management (create, communicate, terminate) - MCP server and tool access control - Console display with agent tree visualization - Usage tracking and performance monitoring Use Cases: - Project management with specialized development teams - Code review teams with different expertise areas - Content creation with writers, editors, and reviewers - Data analysis with collectors, processors, and visualizers
- Organize imports alphabetically and group properly - Remove unused imports (typing.Any, TextContent, PromptMessageMultipart, etc.) - Fix import formatting across all example files - Ensure compliance with project code style guidelines
evalstate
reviewed
Aug 5, 2025
| model: Optional[str] = None | ||
|
|
||
|
|
||
| class DynamicAgentInfo(BaseModel): |
There was a problem hiding this comment.
We already use the AgentCard (from the A2A framework) to describe Agents and Tools for Router/Orchestrator. Would it be possible to standardize on this type? And we have serialization routines for this (I think... check the router/orchestrator)
evalstate
reviewed
Aug 5, 2025
src/mcp_agent/agents/base_agent.py
Outdated
| content=[TextContent(type="text", text=formatted)] | ||
| ) | ||
|
|
||
| elif name == f"{DYNAMIC_AGENT_TOOL_PREFIX}_list": |
There was a problem hiding this comment.
Can we have a think whether this would be best injected as part of the instructions/early messages? Rather than in a Tool Call. The text that describes the available agents doesn't need to be hidden behind a tool call. (Would consider this)
evalstate
reviewed
Aug 5, 2025
src/mcp_agent/agents/base_agent.py
Outdated
| content=[TextContent(type="text", text=response)] | ||
| ) | ||
|
|
||
| elif name == f"{DYNAMIC_AGENT_TOOL_PREFIX}_broadcast": |
There was a problem hiding this comment.
If were super-super MCP minded, we would set up a Resource Subscription for this. (This will add work for little benefit).
- Update dynamic_agent_manager.py to import from fast_agent.agents.agent_types - Update test_dynamic_agent_manager.py to import from fast_agent.agents.agent_types - All 19 dynamic agent tests now pass successfully - Resolves import errors caused by package restructuring in main branch
Addresses feedback to use existing AgentCard type instead of custom DynamicAgentInfo. Changes: - Replace DynamicAgentInfo class with create_dynamic_agent_card() function - Use AgentCard from a2a.types for consistency with router/orchestrator - Convert server access, status, and usage info into AgentSkill objects - Update list_agents() method to return List[AgentCard] instead of List[DynamicAgentInfo] - Update all tests to work with new AgentCard structure - Maintain all original functionality while using standardized types Benefits: - Consistency with existing codebase architecture - Access to existing serialization routines (model_dump_json) - Better integration with router/orchestrator components - Future-proof compatibility with A2A framework All 20 tests pass successfully.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
#326
[https://drive.google.com/file/d/1Hrl2tHw_NfJNRfuuimOctbOh2Co_CjCo/view?usp=sharing](Video Preview)