agent-factory

Agent Factory

Multi-agent orchestration for your AI coding assistant.

Describe what you want — Agent Factory breaks it into stages, assigns agents, and executes the full pipeline. Ships as an MCP server with 27 tools. Works with Claude Code, OpenCode, or any MCP client.

npm install -g agent-factory

That's it. Python dependencies are auto-installed. Restart your AI tool and start using it.

What Can You Do?

Diagnose Bugs

Give it an error message or symptom. It scans your project, finds the root cause, and proposes a fix.

You: "ZFS quota list isn't showing up. Find out why."

 → perform_rca scans your code
 → Returns root cause analysis + fix proposal
 → approve_workflow → auto-generates the patch

Turn Ideas Into Code

Describe a feature vaguely. It asks clarifying questions, refines the requirements, then implements.

You: "I want real-time notifications."

 → refine_idea asks 3-5 clarifying questions
 → You answer → submit_idea_refinement
 → Refined requirements + implementation plan
 → approve_workflow → architecture → code → validation

Run a Full Pipeline

One prompt, end-to-end execution.

You: "Build a REST API with JWT auth and user management."

 → execute_workflow(work_type="software_development")
 → Analysis → Design → Implementation (writes files) → Validation

Install

npm install -g agent-factory

Postinstall automatically:

• Creates Python venv at ~/.agent-factory/venv
• Installs mcp, pyyaml, and the agent-factory package
• Detects your AI tool and installs platform components

Requirements: Node.js ≥ 14, Python ≥ 3.10

No-Install Run

npx agent-factory

Platform Setup

OpenCode — auto-configured during install. Just restart.

Claude Code — register manually:

claude mcp add agent-factory -- npx agent-factory

Any MCP client — add to your config:

{
  "mcpServers": {
    "agent-factory": {
      "command": "npx",
      "args": ["agent-factory"]
    }
  }
}

Verify

Restart your AI tool. You should see tools like execute_workflow, perform_rca, refine_idea in the MCP tools list.

Workflow Types

Type	Stages
software_development	analysis → design → implementation → validation
ml_development	problem → data → design → training → evaluation → deployment
test_generation	analysis → design → validation
data_analysis	problem → data → evaluation
code_review	analysis → validation

Omit work_type to auto-detect.

Documentation

Doc	Content
Setup Guide	Detailed install, env vars, troubleshooting
MCP Tools Reference	All 27 tools — params, returns, examples
Architecture	Internal structure, core components, state
API Reference	Python classes and methods
OpenCode Integration	Plugin structure and setup
OpenCode Usage Guide	Practical usage in OpenCode
Changelog	What's new in each version

License

MIT