Ship with AI Agents That Follow Your Rules
Agiflow connects your agile workflow with AI coding agents — so every generated file follows your patterns, your architecture, and your conventions.
Why AI Agents Break Down at Scale
AI coding agents are powerful for prototypes. But as projects grow, four problems compound:
No conventions
Every file follows different patterns. You spend hours reviewing code that works but doesn’t fit your codebase.
No coordination
Parallel agents create conflicts — duplicate auth libraries, broken API contracts, conflicting migrations.
No memory
Agents lose context between sessions. You re-explain architecture every time. 6 tasks become 6 isolated conversations.
No validation
Violations only surface in code review — after hours of agent work. Missing error handling, wrong file locations, skipped tests.
How Agiflow Solves This
Two complementary layers — one open source, one platform:
Open Source: AI Code Toolkit
MCP servers and CLI tools that enforce how agents write code. Works with Claude Code, Cursor, Windsurf — any MCP-compatible agent.
Generate standardized code from templates. Reduce 2000 lines of boilerplate to 200.
Convention over configuration. Agents call get-file-design-pattern before writing code.
Post-flight review-code-change catches violations with severity-rated feedback.
Agiflow Platform
Transforms your agile workflow (epics, user stories, acceptance criteria) into AI-executable work with coordination and observability.
Epics with tasks, acceptance criteria, and devInfo metadata. Agents resume with full context — no re-explaining.
Launch Claude, Gemini, or any agent with task context. Auto-provisions MCP tools per task requirements.
Specialized agents (backend, frontend, reviewer) with role-based MCP tool access and Docker sandbox isolation.
Live session tracking, tool usage audit, progress visualization, and comment system for agent feedback.
How It Works
A typical workflow from task creation to validated code:
Create a Work Unit with Tasks
Define an epic like "Shopping Cart Feature" with tasks, acceptance criteria, and devInfo metadata (git worktree, Docker container, test baseline). Designed for AI execution, not manual development.
Launch an Agent from the Dashboard
Start a chat for a task. Agiflow reads acceptance criteria, parent context, and related tasks, then provisions the right MCP servers (scaffold-mcp, architect-mcp, database tools) for the agent's role.
Agent Scaffolds and Builds
Agent calls scaffold-mcp to generate boilerplate, then architect-mcp to understand design patterns before adding business logic.
Validate Against Rules
Agent calls review-code-change to catch violations — missing error handling, wrong patterns, skipped tests. Fixes and re-validates until clean.
Review, Comment, Resume
Review in the dashboard. Add feedback as comments. Restart the agent — it reads comments, devInfo, and previous test results. Full context retained, no re-explaining.
Explore the Features
Spec + Task Management
Group tasks into Work Units with persistent devInfo. Agents resume with full context across sessions.
See how Work Units workReliable Agent Execution
Specialized agent roles with session tracking, model routing, and scoped MCP server access per task.
Explore agent rolesSecurity & Governance
RBAC with 4 roles, Docker sandbox isolation, and per-agent MCP tool access controls.
Review security modelMCP Servers
Project MCP, Scaffold MCP, Architect MCP, and One MCP — the open-source toolkit that powers agent coordination.
Browse MCP serversStart with the Open Source Toolkit
No platform required. Add scaffolding templates and architecture validation to your existing workflow in under 5 minutes. Works with Claude Code, Cursor, and any MCP-compatible agent.
Start Building with AI Agents
Give your AI agents the conventions, coordination, and validation they need to ship production code.