Built for AI Agents

Agent workflow engine for AI agents

Define multi-step workflows with validation at every step. Your AI agent executes consistently, every time.

Think of it as CI/CD for AI agent tasks.

claude-code
❯ Start development workflow
Starting moira/software-development-flow-lite...
β–Έ directive: Study project foundations
❯ Project studied. Moving to planning...
β–Έ directive: Create implementation plan
Plan created: 4 steps
❯ Plan approved. Implementing step 1...
β–Έ directive: Implement authentication layer
βœ“ Step 1/4 complete. Tests: 42/42 passed
Workflow: 25% complete
βœ“ MCP Protocol Native
βœ“ Open Source Workflows
βœ“ JSON Schema Validation

The Problem

What happens without structured workflows

🧠

Lost Context

Long prompts that the agent forgets halfway through. Complex tasks break into pieces that never reconnect.

❌

No Result Validation

Agent says "done" but the code doesn't compile. No schema checks, no proof of completion.

πŸ‘»

Hallucinated Results

Agent invents files, functions, and test results that don't exist. Confident but wrong.

⏳

80% Done, 20% Forgotten

Tests skipped. Documentation missing. Error handling ignored. The last mile never gets finished.

πŸ”„

Re-explaining Every Time

Same process, different session. You explain the workflow from scratch because nothing persists.

🎯

No Execution Control

Agent skips steps, jumps ahead, takes shortcuts. No way to enforce a specific execution order.

How It Works

Three steps to structured execution

1

Connect

Add Moira as an MCP server to your AI agent. One config line in Claude Code, Cursor, or any MCP client.

2

Start a Workflow

Choose a workflow β€” built-in or custom. The engine loads the execution graph and begins step-by-step processing.

3

Agent Executes

Each step has a clear directive and completion criteria. The agent works through the graph β€” validated, structured, complete.

What You Get

Built for Real Development Workflows

πŸ€–

Agent Does the Work

From planning to self-review. Define the task β€” get a working result. The workflow guides the agent through every phase: requirements β†’ plan β†’ implementation β†’ tests β†’ quality β†’ review.

πŸ›‘οΈ

Hallucination Protection

JSON Schema validation at every step. The agent cannot skip ahead β€” each response must match the expected structure. Completion conditions define what "done" actually means.

βœ…

Enforced Code Quality

Mandatory code quality checks and agent self-validation. Bounded fix loops β€” if quality doesn't meet standards, the agent retries (up to 3 attempts), then escalates to you.

πŸ§ͺ

Automated Test Validation

Test execution, failure analysis, and fix strategies β€” all inside the workflow. The agent doesn't just write code β€” it verifies that the code works.

πŸ“„

Docs and Tests Stay Current

If your project has documentation and tests, the workflow includes their updates. The completion phase updates docs and generates a final report.

πŸ”€

Branch Isolation Built In

Git workflow is part of the process β€” commits, PR creation, branch management. Everything is isolated from master until reviewed and approved.

πŸ“±

Telegram When You're Needed

You don't sit in front of a screen. When the workflow needs your input β€” approve a plan, review code, resolve an issue β€” you get a Telegram notification.

Production Workflows

Ready-Made Development Workflows

Battle-tested workflows that cover the full software development lifecycle. Start using them today or customize to your needs.

Complete development workflow for features and complex tasks. Covers the entire lifecycle from requirements to deployment.

153
Total Nodes
79
Agent Directives
50
Condition Nodes
9
Telegram Alerts
19
Workflow Variables
1
Requirements
Gather requirements, study project, define task scope
2
Planning
Create plan + agent review + fix cycle (max 3 iterations)
3
User Approval
Review and approve the plan before implementation
4
Implementation
Step-by-step execution with progress tracking
5
Testing
Run tests β†’ analyze failures β†’ fix strategy
6
Code Quality
Quality check + self-validation with bounded fix loops
7
Browser Testing
UI testing in browser (when applicable)
8
User Review
Final review by developer with approval/rejection
9
Completion
Report, documentation update, notifications

Quick Start

Start in 3 Steps

1

Connect MCP Server

Add Moira as an MCP server in your AI client. One URL, OAuth authentication.

# Claude Code (CLI)
claude mcp add moira https://moira.witqq.ru/mcp

# Cursor / VS Code (settings.json)
{
  "mcpServers": {
    "moira": {
      "url": "https://moira.witqq.ru/mcp"
    }
  }
}
2

Start a Workflow

Tell your AI agent to start any workflow. Moira handles the rest.

# In any MCP-compatible AI client:
> Start the software development workflow
  for issue #42

# Agent calls:
mcp__moira__start({
  workflowId: "moira/software-development-flow",
  parentExecutionId: "none"
})
3

Follow the Workflow

Agent receives step-by-step directives with success criteria. Each step validated before proceeding.

# Moira returns:
{
  "directive": "Implement the feature
    according to the approved plan",
  "completionCondition": "All tests pass,
    code follows project patterns",
  "inputSchema": { ... }
}

# Agent works β†’ submits result
mcp__moira__step({
  processId: "abc-123",
  input: { ... }
})

Extensible

Build Your Own Workflows

Ready-made workflows are the starting point. Create custom workflows for your team's specific processes.

πŸ–₯️

Visual Web Editor

Create and edit workflows through the web interface. Node-graph editor with real-time validation.

πŸ“„

JSON-first Design

Workflows are JSON files. Version control, code review, CI/CD β€” standard developer workflow applies.

πŸ”—

Template Variables

Use {{variable}} syntax in directives and conditions. Context flows between steps automatically.

πŸ›‘οΈ

Schema Validation

JSON Schema on every step input/output. Agent responses validated before workflow proceeds.

Real-world Applications

What Developers Build

πŸš€

Automated Feature Development

Full cycle: issue analysis β†’ planning β†’ implementation β†’ testing β†’ code review β†’ commit. With quality gates at each phase.

πŸ”

Structured Code Review

Automated review with BLOCKING/MAJOR/MINOR classification. Consistent criteria across all PRs, no reviewer fatigue.

πŸ›

Bug Fixing with Root Cause Analysis

Reproduce β†’ diagnose root cause β†’ implement fix β†’ verify no regression. Prevents band-aid fixes.

πŸ“š

Documentation Generation

Extract API docs from code, generate guides from usage patterns, keep docs in sync with implementation.

πŸ”¬

Research with Iterative Improvement

Source verification, cross-referencing, quality scoring. Multiple iterations until research meets quality threshold.

What Developers Say

Built by developers, for developers

"Moira turned our chaotic prompt engineering into repeatable workflows. Code review consistency went from random to reliable."

A

Alex K.

Senior Backend Engineer

"The JSON-first approach means I can version control my workflows. Schema validation catches agent mistakes before they reach production."

S

Sarah M.

DevOps Lead

"Three steps to get started, and it actually works. No sign-up wall, no sales call. Just MCP connect and go."

C

Chen W.

Full Stack Developer

Ready to Build?

Stop micromanaging your AI agent. Define the workflow once, execute it consistently every time.