MCP Moira
Integrates with GitHub through the account's server-held GitHub OAuth/App authorization, allowing Moira to enroll and work with private Git repositories (including write operations) while running approved repositories in its local Docker Sandboxes.
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 and its ingress through Docker Compose 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: Heym
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/local/ - Paired controller for workspace provisioning and optional Docker Sandboxes microVMs
packages/workspace-service/ - Common guest command, file and session service with its Unix-socket CLI
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
Compose starts the application and static Traefik ingress on the same public port. Keep
config/traefik-self-host.yml with the Compose file; the Self-Hosting guide
includes the three-file download procedure. The ingress forwards /api/workspaces/ HTTP/1.1 Upgrade
directly to backend port 3001 and ordinary traffic to nginx 80, without Docker socket access.
Its application TRUST_PROXY default is 2; configure explicit trusted proxy addresses for another
ingress topology. A standalone frontend-port mapping is not the supported workspace stream route.
The image is pulled from the public registry by default. Cloud account, workflow and execution
data persists in ./data. Workspace files, operation results and immutable file publications stay
in the workspace service or paired controller's private storage. Back up those owner stores with
their VM disks; a cloud database backup does not contain their file bytes.
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 ordinary text-channel handlers run in the separate extension runner and are disabled by default. File-capable channels additionally declare a fixed binaryTransport and run their streaming binaryHandler beside the source at its workspace or paired controller. Enabling extensions
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.
Local codespaces on your computer
Moira Local connects a paired controller to your Moira account through outbound HTTPS. The controller can provision the common service in GitHub Codespaces and can run locally approved repositories in persistent, mountless Docker Sandboxes microVMs. Enrollment and controller startup require no Local VM initialization or Docker sign-in. Public read-only Local work needs no GitHub App connection; private Git and write operations use repository-scoped credentials derived from the account's server-held GitHub authorization.
The controller requires Node.js 24. Its optional physical Local VM capability requires macOS on
Apple silicon and the pinned Docker Sandboxes sbx 0.46.0 runtime; Linux Local VM execution is
refused. From this source checkout:
npm ci
npm run local:build
npm run local -- --helpThe package is private and is not a published global npm installer. Pair the controller through
Settings → Development → Local computers; the pairing token reaches the companion through stdin.
For Local VM work, initialize the physical capability and approve repository rights and a finite
work lease on that computer. Review effective owner grants in the browser before remote work.
See Codespaces: Moira Local
for setup, run, stop, recovery and the limits of isolation evidence. Codespace contents are visible
to Moira; review returned code before executing it on your computer. The moira-local MCP connection
name in the examples below is independent of the companion executable.
The same workspace service owns commands, files, sessions and output in Local and GitHub guests.
The Local owner prepares its guest; the paired controller performs fixed GitHub provisioning or
explicit in-place adoption. The service includes its own Node.js 24.21.0 and ripgrep 15.2.0, independent of
project Node/PATH. Inside a prepared workspace, ~/.local/bin/moira-workspace --help exposes the
direct Unix-socket CLI. Choose ws1-<lowercase UUID> for command and mutation requests; plain UUIDs
remain repository-admission identities. Inspect the original identity after a lost reply before
repeating an effect. Command output and file bytes stream without Moira byte/concurrency/runtime
quotas; explicit observation/deadline options, real storage capacity and repository/owner guards
remain. A ready published attachment can be sent through
the separately requested communication action deliver-file. See
File operations and native transfer
for current authority, link expiry and guarded target replacement. The common OCI image is
ghcr.io/moira-mcp/moira-workspace-service; a project keeping its own Linux Dev Container can use
the optional ghcr.io/moira-mcp/features/workspace-service Feature with a published release tag.
Host moira-local create, start, stop and remove call the running owner controller with the
observed generation and an original request ID; collect inspects that ID without repeating work.
Creation also needs cloud route registration and initial Git authority. Existing owner-admitted
Local work and its guest CLI remain subject to current local grants and leases during a temporary
cloud outage. A confirmed authority refusal does not grant offline access. Raw host request and
credential-export commands are unsupported; recover remains an explicit custody action.
For GitHub, a repository without the workspace Feature needs an online assigned controller for fixed installation. An existing environment uses explicit adoption without replacement; a checked-in Feature is an opt-in startup path. Guest readiness requires authenticated outbound registration. A definitive registration refusal fails startup, while temporary network loss reconnects observation without replaying commands or lifecycle operations.
Workflow materialization resolves authorized text in Moira and creates its UTF-8 tar at a selected
paired controller. The download grant pins that controller and original publication request; repeated
reads collect the same immutable archive without rendering again or choosing another owner. If the
first publication is unconfirmed, collect that grant or request a fresh current-step presentation for
an explicit retry. The in-context session materialize fallback requires no file owner and retains
its separate response-size budget.
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
Account API routes require authenticated access; public capability routes enforce their own narrower grants
The guest workspace's outbound connection receives authority for that resource and measured environment only; it receives no full MCP, GitHub or communication credential
Shared middleware and service guards enforce the required account and resource authority
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": "deliver-file", "request_id": "ws1-11111111-1111-4111-8111-111111111111", "referenceId": "<file.reference_id>", "message": "Report", "kind": "document"}
communication {"action": "operation", "request_id": "ws1-11111111-1111-4111-8111-111111111111", "device_id": "<device_id returned for that delivery>"}
# Documentation
help
help {"topic": "tools"}
help {"topic": "step"}File delivery requires a live source owner and a caller-chosen original request_id.
Keep the returned owner identity and collect that same request after uncertainty; for a
workspace owner, use codespace_id instead of device_id. Collection never sends again.
Owner publications contain immutable bytes and support repeated reads until expiry under
current authority. Ordinary text, waiting reminders and trusted lock PIN delivery need no
file owner. Optional workflow progress images can be omitted while text sends independently;
an explicit file delivery cannot silently succeed as text. Progress-image mint/read reports
COMMUNICATION_OWNER_UNAVAILABLE with HTTP 503 when its paired controller is unavailable.
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:
Ordinary API routes: 100 requests/minute
Auth routes (
/api/auth/*): 1000 requests/minuteMCP endpoint (
/mcp): 1000 requests/minuteWorkspace channel: resource-scoped authorization is checked separately; binary streams do not pass through the ordinary JSON API body parser or its rate-limit queue
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
Administrator Overview
The /admin page shows deployment-neutral system health and managed-workflow reconciliation.
The Cloud adminAnalytics capability also enables recently active people, registrations,
executions needing attention, popular flows and execution analytics. Each card has its own period.
Activity means a recorded action or refreshed unexpired session, not continuous online presence.
Analytics excludes current administrators by default. Individual exclusions and an explicit include-everyone choice are saved separately for each account in the current browser. Management lists and system health remain outside these exclusions; default self-host does not expose these installation-wide analytics.
User Management
Admin panel at /admin/users provides:
User list with approval, email verification, and blocked status
Work context with recorded activity, accepted-step time, run count and most-used flows; unknown activity or step time is identified explicitly
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, workflow ID or note
Read owner, note, lock/refusal badges, known start time and the last accepted step separately; a record edit does not change step time, and missing timestamps are not fabricated
Open permitted run details, including recorded progress, locks and errors
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, the overview of runs, 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 and local providers, companion setup/recovery, persistent lifecycle, MCP tools, website management and 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.
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
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
- 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.209MIT
- 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.21,183-
- FlicenseAqualityDmaintenanceAn agent-native workflow MCP server that enables AI agents to execute text-defined, versionable workflows with checkpointing and state management.107 npm-
- AlicenseAqualityAmaintenanceA local, auditable multi-model workflow engine that lets you define YAML graphs for orchestrating LLM agents across vendors, with MCP tools for validation, dry-runs, execution, and human approval, all fully observable in a local web interface.83Apache 2.0