MCP Moira
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@MCP Moirarun the invoice-approval workflow for invoice 1042 and tell me which branch it took"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
MCP Moira
Agent Workflow Engine for AI agents.
Primary users: AI agents via MCP protocol. Web UI is supplementary for workflow management.
See docs/VISION.md for product vision and design principles.
Two ways to run Moira
Self-host (this repository, Apache-2.0) — run the full engine + Web UI + MCP server in a single Docker container on your own infrastructure. Free and open source; your data stays with you. Start in the Quick Start below.
Moira Cloud (managed) — a hosted instance with nothing to operate, at moira-mcp.com.
Both run the same engine and MCP tools. Self-host is a single-tenant private-team deployment with administrator-approved accounts. Cloud adds managed hosting and SaaS-only policy and administration, including social login, legal consent, email verification, and the broader multi-user administration surface.
Related MCP server: task-orchestrator
Architecture
Monorepo: Clean separation of concerns with npm workspaces Workflow Engine: Node-graph execution over a set of node types (@mcp-moira/workflow-engine) MCP Server: HTTP protocol server exposing the MCP tools (@mcp-moira/mcp-server) Web Backend: Express API server (@mcp-moira/web-backend) Web Frontend: React UI with webpack (@mcp-moira/web-frontend) Docs: Astro 5 + Starlight documentation site, EN+RU (@mcp-moira/docs) Shared: Database layer + Better Auth + logging (@mcp-moira/shared) Database: Modular repository pattern with Drizzle ORM Settings System: Universal settings with encryption and dynamic UI generation Docker Deployment: Multi-stage container with TypeScript validation Validation: JSON Schema with AJV
Package Structure
packages/workflow-engine/ - Core node-graph execution engine
packages/extension-sdk/ - Typed authoring contract for custom nodes and communication channels
packages/extension-runner/ - Isolated companion service for installed extension bundles
packages/mcp-server/ - MCP protocol HTTP server with tools
packages/web-backend/ - Express API for workflow management
packages/web-frontend/ - React UI for workflow visualization
packages/docs/ - Astro 5 + Starlight documentation site (EN+RU), built into the image and served at
/docspackages/shared/ - Database layer (schema, connection, repositories) + Better Auth + logging
database/- Modular repositories (Workflow, Execution, Settings)auth/- Better Auth configurationlogging/- Structured logging
Docker Config (
config/) - Unified container deployment configuration
Quick Start
Self-Host (recommended)
Run a complete Moira instance locally with Docker — no source build required:
cp .env.example .env # defaults work locally; review host, port, and artifact domain for another host
docker compose up -dThen open:
Web UI: http://localhost:8080
Documentation: http://localhost:8080/docs/
MCP endpoint: http://localhost:8080/mcp
The image is pulled from the public registry by default. Data (SQLite + execution
storage) persists in ./data. See Self-Hosting or the in-app docs
at /docs/ for the full reference.
Updating / Upgrading
Update a normal self-host installation with the standard Compose commands:
docker compose pull
docker compose up -d
docker compose psOlder .env files may still override Compose with the removed 0.3.5 tag; change that line once to
MOIRA_IMAGE=ghcr.io/moira-mcp/moira:latest before updating.
The image protects existing self-host data before its startup migrations: it creates and verifies a
coherent SQLite backup, includes the prompt manifest, and keeps three rotating recovery states under
data/.moira-startup-backups/. If initialization fails, it restores the database and manifest before
refusing to start the services. A persistent pending marker also restores the verified state before the
next attempt if the container or host was interrupted mid-initialization. A fresh installation skips
the nonexistent-database backup and uses only a temporary persistent marker so an interrupted first
start is removed before retry.
The complete automatic recovery behavior and optional pinned-image preflight are documented in Self-hosting: Updating and Recovery, with a matching Russian version. Release notes are on the GitHub Releases page.
Local Development (from source)
For contributors who want to build and run from the source tree, switch
docker-compose.yml to Option B first — comment out the image: line and
uncomment the build: block (the file documents both options inline). Then:
nvm use
npm install
docker compose up -d --build # builds the image locally from config/Dockerfile
# Web UI: http://localhost:8080 | MCP: http://localhost:8080/mcp(The default docker-compose.yml uses the prebuilt public image — docker compose up -d without --build — which is the recommended self-host path.)
Extensions (optional)
Custom node and communication-channel bundles run in the separate extension runner and are disabled by default. Enabling them requires this source checkout because the runner image is built locally; the published Moira image does not import or package bundle code.
mkdir -p extensions
cp -R examples/extensions/webhook-notify extensions/
printf '\nMOIRA_EXTENSION_RUNNER_URL=http://moira-extension-runner:9110\n' >> .env
docker compose --profile extensions up -d --buildBefore starting the example, replace both placeholder network permissions and fill its per-user
settings. The example contributes both a workflow action node and an ordinary notification channel.
See Self-hosting: Enable
extensions
and Writing an
Extension for the complete
installation, SDK, permission and failure contracts. npm run test:docker-extensions verifies the
default-off profile and read-only bundle mount.
Testing
The integration/API/E2E suites run against a local Docker container, configured by
.env.local. Copy the template once before running them (or before
npm run docker:restart):
cp .env.local.example .env.local # then set BETTER_AUTH_SECRET
npm test # All tests
npm run test:unit # Unit tests only (no container needed)
npm run test:e2e # E2E testsCode Quality
npm run fix # ESLint + Prettier fix all filesConfiguration in .env (copy from .env.example):
MOIRA_PORT: External access port (default 8080)
MOIRA_HOST: Public host:port the instance is served on (default localhost:8080)
STATIC_ARTIFACTS_DOMAIN: Wildcard subdomain base for published artifacts
BETTER_AUTH_SECRET: generated and persisted on first start when empty; set it explicitly only when you want to manage the auth signing secret yourself
Database: SQLite at
./data/moira.db(bind-mounted, persists across restarts)Admin: ADMIN_EMAIL, ADMIN_PASSWORD (auto-generated on first start if unset)
Authentication
MCP Moira uses Better Auth with OAuth 2.1 for centralized authentication.
Browser Access:
Email/password login at http://localhost:8080/login
GitHub/Google OAuth (saas mode only; disabled in self-host)
Better Auth UI components (Tailwind + shadcn/ui)
MCP Clients:
OAuth 2.1 authorization code flow
HTTP 401 triggers OAuth discovery
Dynamic Client Registration (DCR) supported
Access token required for all MCP tool calls
Protected:
All MCP tools require authentication
All API routes (/api/) require authentication (except /api/auth/)
Centralized protection via middleware (no manual checks)
Testing:
docker compose up -d
# Access: http://localhost:8080/login
# MCP Inspector: http://localhost:8080/mcpSee docs/AUTHENTICATION.md for complete setup and OAuth flow details.
MCP Configuration
Point your MCP client (e.g. Claude Code) at your running instance:
Server | URL | Purpose |
| Your local self-host instance |
{
"mcpServers": {
"moira-local": { "url": "http://localhost:8080/mcp" }
}
}Replace
localhost:8080with your own host/port (MOIRA_HOST) if you serve Moira on a different address.
Representative Node Examples
The examples below show common graph patterns; they are not the complete node-type inventory. See the Nodes reference for every supported type and its current contract, including automatic note operations and file materialization.
Start Node
{
"type": "start",
"id": "start",
"connections": { "default": "next-node-id" }
}Agent Directive Node
{
"type": "agent-directive",
"id": "task",
"directive": "Task instruction",
"completionCondition": "Success criteria",
"inputSchema": {/* JSON Schema */},
"connections": { "success": "next-node-id" }
}Condition Node
{
"type": "condition",
"id": "check",
"cases": [
{
"when": { "operator": "gte", "left": { "contextPath": "score" }, "right": 8 },
"output": "passed"
}
],
"connections": {
"passed": "success-path",
"default": "failure-path"
}
}cases are evaluated in authored order; the first case whose when holds selects its output,
and default is taken when none holds.
User Notification Node
{
"type": "user-notification",
"id": "notify",
"message": "Task completed: {{result}}",
"format": "plain",
"connections": { "default": "next-node-id" }
}The node fans out to the current user's enabled communication channels and never accepts provider,
recipient, or credential fields. The deprecated telegram-notification node remains available for
existing Telegram-specific workflows, including exact explicit chatId semantics.
End Node
{
"type": "end",
"id": "end",
"finalOutput": ["result", "score"]
}Expression Node
{
"type": "expression",
"id": "increment-counter",
"expressions": ["counter = counter + 1"],
"connections": { "default": "next-step" }
}Teleport Node
{
"type": "teleport",
"id": "teleport-replan",
"directive": "Rewrite the development plan",
"completionCondition": "New plan created",
"hint": "Use when plan needs restructuring",
"connections": { "success": "plan-node" }
}Subgraph Node
{
"type": "subgraph",
"id": "run-subtask",
"graphId": "subtask-workflow",
"inputMapping": { "parentVar": "subVar" },
"outputMapping": { "subResult": "parentResult" },
"connections": { "success": "next-step" }
}Lock Node
{
"type": "lock",
"id": "approval-gate",
"reason": "Waiting for user approval before deployment",
"connections": { "unlocked": "next-step" }
}Pauses execution until explicitly unlocked. Sends PIN via Telegram with inline approve button. Unlockable via MCP tool, web UI, or Telegram callback.
Workflow Format
{
"id": "workflow-id",
"metadata": {
"name": "Workflow Name",
"version": "1.0.0",
"description": "What this workflow does"
},
"nodes": [/* Node definitions */]
}Templates
Variables processed in directive, completionCondition, and message fields:
{{variable}}- Context variable{{nested.path}}- Object property access{{executionId}}- System: current process ID{{workflowId}}- System: current workflow ID{{runUrl}}- System: the run's page in the web app
MCP Tools
# Workflow Management
list
start {"workflowId": "workflow-id"}
step {"processId": "process-id", "input": "data"}
manage {"action": "create", "workflow": {...}}
manage {"action": "edit", "workflowId": "workflow-id", "changes": {...}}
manage {"action": "get", "workflowId": "workflow-id"}
# Session Information
session {"action": "user"}
session {"action": "executions"}
session {"action": "execution_context", "executionId": "execution-id"}
session {"action": "current_step", "executionId": "execution-id"}
# Execution Locking
lock {"action": "lock", "executionId": "execution-id", "reason": "Awaiting approval"}
lock {"action": "unlock", "executionId": "execution-id", "pin": "123456"}
lock {"action": "status", "executionId": "execution-id"}
lock {"action": "list"}
# User Settings
settings {"action": "get"}
settings {"action": "get", "category": "codespaces"}
settings {"action": "set", "key": "codespaces.idle_timeout_minutes", "value": 60}
settings {"action": "list"}
# Workflow Tokens
token {"action": "upload", "ttlMinutes": 60}
token {"action": "download", "workflowId": "workflow-id", "ttlMinutes": 60}
# User Communication
communication {"action": "send", "message": "The report is ready."}
communication {"action": "attachment-token", "message": "Report", "kind": "document", "filename": "report.pdf", "mimeType": "application/pdf", "sizeBytes": 12000}
# Documentation
help
help {"topic": "tools"}
help {"topic": "step"}File Structure
packages/workflow-engine/ # Core execution engine
packages/mcp-server/ # MCP HTTP server (internal port, behind nginx)
packages/web-backend/ # Express API (internal port, behind nginx)
packages/web-frontend/ # React UI (static build served by nginx)
data/ # SQLite database (moira.db)
workflows/ # Bundled public workflow catalog
docs/ # Technical documentationDevelopment
All development happens through Docker containers.
docker compose up -d --build # Build and run the container
npm test # Run all tests
npm run fix # ESLint + Prettier fixDatabase: SQLite at DB_PATH (default: ./data/moira.db)
Migrations: Drizzle ORM (npx tsx scripts/run-migrations.ts)
Storage: Workflows and executions in database with user isolation
Documentation
User Documentation: Served by your running instance at /docs/ (EN) and /ru/docs/ (RU), built from packages/docs (Starlight).
Technical Documentation: /docs directory - system reference, API specs, development guides.
Project Checklist - mandatory pre-commit checks executed by development workflows.
Claude Code Commands
Custom slash commands in /commands directory. See commands/README.md for installation and usage.
HTTP Transport
Architecture
Stateless Mode: Each HTTP request creates new transport, no session storage
JSON-RPC 2.0: MCP protocol over HTTP with proper error handling
Direct Tools: MCP tools integrated in single process
Environment Inheritance: HTTP server environment variables passed to tools
Endpoints
POST /mcp # JSON-RPC requests (tools calls)
GET /health # Server health checkEnvironment Variables
# Required for Telegram integration
TELEGRAM_BOT_TOKEN=your_bot_token
TELEGRAM_DEFAULT_CHAT_ID=your_chat_id
# Optional
MCP_PORT=4202 # Internal MCP server port (accessed via nginx proxy)
LOG_LEVEL=info
DEBUG_CONSOLE=true # Console logging for development
WORKFLOWS_DIR=./packages/web-backend/workflows/productionConfiguration
MCP server configuration (see MCP Configuration for details):
{
"mcpServers": {
"moira-local": {
"url": "http://localhost:8080/mcp",
"type": "http"
}
}
}Environment variables passed via HTTP headers (recommended for HTTP transport).
Alternative: Set environment variables in your .env file.
Code Quality
ESLint + Prettier
Project uses ESLint with TypeScript support and Prettier for code formatting.
npm run fix # Auto-fix lint errors and format codePre-commit Hook: Husky pre-commit hook automatically runs ESLint and Prettier on staged files.
Configuration:
.eslintrc.json- ESLint rules (strict for production code, relaxed for tests).prettierrc- Prettier formatting rulesProduction code:
anytypes are errors, must be properly typedTest code:
anytypes allowed for flexibility
Security
Rate Limiting
Protection against spam and DoS attacks with tiered limits:
API routes (
/api/*): 100 requests/minuteAuth routes (
/api/auth/*): 100 requests/minuteMCP endpoint (
/mcp): 30 requests/minute
Exceeded limits return HTTP 429 Too Many Requests.
Data Size Limits
Protection against oversized payloads:
Workflow JSON: max 5MB
Execution context: max 10MB
Exceeded limits return HTTP 413 Payload Too Large.
GeoIP Logging
Request logging includes country detection via geoip-lite:
{
"method": "POST",
"path": "/api/workflows",
"ip": "203.0.113.1",
"country": "US",
"duration": 45,
"status": 200
}Admin Features
User Management
Admin panel at /admin/users provides:
User list with approval, email verification, and blocked status
User details page with sessions, OAuth connections, email history
Approve pending self-host registrations
Session management - revoke individual sessions or all sessions
OAuth management - revoke tokens by provider or all OAuth connections
Block/Unblock users with reason
Send verification email manually
Send password reset email manually
Set a temporary password for an ordinary user when email delivery is unavailable
Execution Monitoring
The Cloud multiUserAdmin capability enables the cross-user panel at /admin/executions. It is
server-denied and hidden by the default self-host policy:
View all user executions
Filter by user, status
Search by execution ID or workflow ID
Inspect execution context and variables
Email History
Track every email attempt:
Verification emails
Password reset emails
Notifications
Status (
sent,failed, or log-onlylogged) with error messages
Email Features
Email Verification (SaaS)
Verification email sent on SaaS signup; self-host registration uses administrator approval instead
Link expires in 24 hours
Admin can resend manually
Password Reset
With a real SMTP or Brevo provider, the user requests via
/forgot-passwordand receives a reset linkLink expires in 1 hour
Without real delivery, the reset form and email-send actions report the capability as unavailable; an administrator can set a temporary password for an ordinary user and require replacement at the next login
Email Provider
Configured via environment variables:
EMAIL_PROVIDER=smtp # smtp, brevo, auto, none, or explicit test sink
EMAIL_FROM=noreply@domain # Required for real delivery
SMTP_HOST=smtp.example.com
SMTP_PORT=587
SMTP_REQUIRE_TLS=true
# SMTP_USER and SMTP_PASSWORD are optional, but must be supplied togetherReal delivery supports generic SMTP and Brevo. With EMAIL_PROVIDER=auto (or
unset), a complete SMTP configuration takes precedence and a legacy
BREVO_API_KEY plus EMAIL_FROM configuration remains supported. The explicit
test provider logs messages and is never advertised as real delivery.
Documentation Map
Where things are documented. After changing code, find the area below and update the matching file in the same change.
Public docs — MDX shells plus MCP-owned portable help
Rendered pages live under packages/docs/src/content/docs/docs/ (EN) and
packages/docs/src/content/docs/ru/docs/ (RU). Runtime-visible topics keep their semantic Markdown
under packages/mcp-server/src/help/content/, with Russian counterparts under …/content/ru/.
Each localized MDX shell imports its matching source through the MCP package. The help tool
discovers and reads the English sources directly; pages with an interactive insertion compose the
same authored before/after sections in runtime and public presentation. The special tools topic is
listed alongside those file-backed topics and renders directly from the typed MCP contract.
Area | Covers | Path |
Getting started | Introduction (agent-first), quickstart, tutorial on the learning flows, which ready flow to use, self-hosting |
|
Concepts | Workflows, nodes, templates, notes, playbooks, artifacts |
|
Guides | Writing directives, creating & editing workflows, reading a flow and a run, extensions |
|
Reference | Tools, input schema, magic variables, condition operators, validation, workflow catalog |
|
Integration | MCP clients, Claude Code, agent guide, Telegram setup, troubleshooting |
|
Patterns | Branching, validation loop, escalation, subagent review, workspace, and more |
|
Internal docs — docs/
For contributors working on the codebase (implementation detail, not end-user docs).
File | Covers | Path |
Development setup | Build, Docker, local dev, project structure |
|
Testing | Test types, runner, fixtures, antipatterns |
|
API | Backend & admin HTTP API reference |
|
System architecture | Engine, storage, MCP transport, handlers, validation |
|
Authentication | Better Auth, OAuth 2.1, API tokens |
|
Web UI | Frontend architecture, components |
|
Audit system | Audit logging design |
|
Workflows | Workflow authoring, tools, catalog |
|
Design system | UI design tokens and components |
|
Documentation style | How to write internal and public docs |
|
Logging | Structured logging conventions |
|
Codespaces | GitHub authorization, persistent lifecycle, MCP tools, website management, readiness/metrics/kill switches, isolation |
|
Issue management | GitHub issue conventions |
|
Architecture decisions | ADRs (licensing, OSS model, …) |
|
Deployment | Environment variables, restart procedures |
|
Legal | License/legal notes |
|
Contributing
See CONTRIBUTING.md for setup, the PR flow, DCO sign-off, and how releases are automated (Conventional Commits → semantic-release → versioned GHCR image). For upgrading a self-host instance, see Updating / Upgrading.
License
This server cannot be deployed
Maintenance
Related MCP Connectors
Open-source Zapier/n8n alternative as an MCP server: agents build, run and debug your workflows.
Hosted AI agents and workflows with app OAuth, human approval gates, and a run ledger.
Create, browse, remix, collaborate on, and run durable AI workflow nodes from MCP hosts.
Zero-setup MCP gateway securely connecting AI to your tools with authentication and workflows
Related MCP Servers
- AlicenseAqualityAmaintenanceDurable, agent-native AI runtime with native MCP client and server support. Rust core for performance with Python SDK for workflow authoring. Features graph-based workflows, durable execution, A2A protocol support, and multi-agent coordination.822Apache 2.0
- AlicenseNot gradedqualityAmaintenanceServer-enforced workflow discipline for AI agents. An MCP server providing persistent work items, dependency graphs, quality gates, and actor attribution. Schemas define what agents must produce — the server blocks the call if they don't. Works with any MCP-compatible client.207MIT
- FlicenseCqualityAmaintenanceSelf-hosted, source-available AI workflow automation platform. Build multi-agent, RAG, and tool-using pipelines on a visual canvas and publish any workflow as an MCP server (stdio/SSE/Streamable HTTP). Also an MCP client via the agent node.21,292-
- FlicenseAqualityBmaintenanceAn agent-native workflow MCP server that enables AI agents to execute text-defined, versionable workflows with checkpointing and state management.1011 npm-