Skip to content

CLAUDE.md

Latest commit

 

History

History
76 lines (56 loc) · 2.88 KB

CLAUDE.md

File metadata and controls

76 lines (56 loc) · 2.88 KB

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Development Commands

Dependencies & Environment

  • Use uv for dependency management (preferred over pip)
  • Install dependencies: uv pip install -e . or uv pip install -r requirements.txt
  • Install dev dependencies: uv pip install -e ".[dev]"
  • Upgrade agno: uv pip install --upgrade agno
  • Test Python imports: uv run python -c "import agno; print('Agno imported successfully')"

Code Quality

  • Linting: ruff check . --fix
  • Formatting: black .
  • Type checking: mypy .
  • Testing: pytest

Running the Server

  • Direct execution: uv run python main.py
  • Using uv: uv run mcp-server-mas-sequential-thinking
  • Package execution: uvx mcp-server-mas-sequential-thinking

Project Architecture

This is a Multi-Agent System (MAS) for sequential thinking built with the Agno framework and served via MCP.

Core Components

Main Entry Point: main.py contains all core logic:

  • FastMCP server setup
  • ThoughtData Pydantic model for input validation
  • Multi-agent team creation and coordination
  • Sequential thinking tool implementation

Agent Architecture:

  • Team Coordinator: Uses Agno's Team object in coordinate mode
  • Specialist Agents: Planner, Researcher, Analyzer, Critic, Synthesizer
  • Agent Flow: Coordinator receives thoughts → delegates to specialists → synthesizes responses

Key Functions

create_sequential_thinking_team(): Instantiates the multi-agent team with specialized roles sequentialthinking tool: Core MCP tool that processes ThoughtData objects get_model_config(): Configures LLM providers (DeepSeek, Groq, OpenRouter)

Configuration

Environment variables control behavior:

  • LLM_PROVIDER: Provider selection (deepseek, groq, openrouter)
  • {PROVIDER}_API_KEY: API keys for each provider
  • {PROVIDER}_{TEAM|AGENT}_MODEL_ID: Model selection for coordinator vs specialists
  • EXA_API_KEY: For research capabilities

Data Flow

  1. External LLM calls sequentialthinking tool with ThoughtData
  2. Tool validates input via Pydantic model
  3. Coordinator analyzes thought and delegates to relevant specialists
  4. Specialists process sub-tasks using their tools (ThinkingTools, ExaTools)
  5. Coordinator synthesizes responses and returns guidance
  6. Process continues with revisions/branches as needed

Memory & State

  • SessionMemory: In-memory storage for thought history and branches
  • Logging: Structured logging to ~/.sequential_thinking/logs/
  • Branch Management: Supports non-linear thinking with branch tracking

Important Notes

  • This is a high-token-usage system due to multi-agent architecture
  • All agent definitions are contained in main.py
  • The system supports revisions and branching for complex problem-solving
  • Configuration is entirely environment-based (no config files)