convez.ai
MCP ServerAutomaçãoAvançado
ahmet-ozel

Plataforma Atlassian AI Workflow

ahmet-ozel/atlassian-ai-workflow-platform

Automação de IA multi-departamento para Atlassian com workflows duráveis

Ver no GitHub
InicialPluginScore · de 0 a 100

MCP server que automatiza fluxos de trabalho em Jira, Confluence e Bitbucket usando agentes LLM e workflows durádouros do Temporal. Inclui gateway MCP stateless, dashboard administrativo, interface Streamlit e orquestração completa em Docker Compose, com suporte a multi-tenant e credenciais em Vault.

Instalação

$npx convez-ai ahmet-ozel/atlassian-ai-workflow-platform

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 4 meses
Versão
—
Issues abertas
1
Contribuidores
1

Casos de uso

  • Automatizar triagem e atribuição de issues Jira com análise LLM
  • Gerar pull requests em Bitbucket baseado em decisões de IA
  • Publicar documentação automaticamente em Confluence
  • Orquestrar workflows complexos entre múltiplas aplicações Atlassian
  • Governança de automação com dashboard administrativo

Veredito

Solução enterprise-grade para automação Atlassian com arquitetura sólida (Temporal, Vault, MCP stateless). Exige experiência significativa com Docker, workflows distribuídos e ecossistema Atlassian; valor real em ambientes multi-departamento com governance robusta.

README

A tradução para o português deste README ainda está na fila.

Atlassian AI Workflow Platform

Self-hosted, multi-department AI automation for Jira, Confluence and Bitbucket. Assign a Jira issue to the bot and the platform turns it into real work: it reasons over the task with an LLM, makes code changes on a runner, opens a draft Bitbucket pull request, publishes a Confluence page, and comments back on the issue - all driven by durable Temporal workflows.

It ships as a single Docker Compose stack: a stateless Atlassian MCP gateway, a webhook intake service, Temporal workers, an admin dashboard for governance, and a Streamlit chat front end for end users.

Keywords: Atlassian automation, Jira bot, Confluence automation, Bitbucket pull request automation, MCP (Model Context Protocol) server, Temporal workflows, LLM agent, self-hosted, Docker Compose.

Highlights

  • Stateless, multi-tenant MCP gateway - every request carries its own Atlassian credentials in headers, so a single gateway serves many users and departments without server-side sessions.
  • Secrets in Vault - department credentials and tokens are stored in HashiCorp Vault and referenced by handle; raw tokens never live in config or source.
  • Admin dashboard - governance surface for departments, services, capabilities, workflows and pull-request review.
  • Streamlit front end - the end-user surface for credentials, chat and task creation.
  • Temporal automation - durable workflows turn a Jira event into an LLM decision, code changes on a runner, a draft pull request, a Confluence page and a Jira update.
  • Env-driven host ports - every published port comes from infra/.env, so a deployment can avoid clashes without touching Compose or source.

Quick Start

For a first-time setup we recommend bootstrap-only mode: only the admin-dashboard and its dependencies start; every other service is activated on demand, under operator control, from the Setup Wizard inside the dashboard.

cd platform
# infra/.env carries the bootstrap values (Postgres, Vault, host ports, LLM).
# It is git-ignored, so create it on a fresh clone before the first boot.
make boot               # starts postgres + vault + admin-dashboard-{api,ui}

Then open the dashboard in your browser:

http://localhost:33000

The Setup Wizard brings up the remaining services step by step and, on the final step, asks you to add the first department.

Full walkthrough: docs/runbooks/getting-started.md.

If make is not installed (the Windows default):

.\scripts\up.ps1 boot

or, on a POSIX shell:

./scripts/up.sh boot

Published host ports

Every service always listens on a fixed container port internally; only the host-published port is configurable, so a deployment can avoid clashes with other software on the same machine. All host ports are defined once in infra/.env (*_HOST_PORT variables) - change a value there and re-run docker compose up -d. Never edit the Compose file or application source.

ServiceHost portEnv variable
Admin Dashboard (UI)33000ADMIN_DASHBOARD_UI_HOST_PORT
Admin Dashboard API38082ADMIN_DASHBOARD_API_HOST_PORT
Streamlit (chat)38501STREAMLIT_HOST_PORT
Atlassian MCP38090ATLASSIAN_MCP_HOST_PORT
assistant-service38081ASSISTANT_SERVICE_HOST_PORT
automation-service38084AUTOMATION_SERVICE_HOST_PORT
task-intake-service38083TASK_INTAKE_HOST_PORT
Postgres35432POSTGRES_HOST_PORT
Redis36379REDIS_HOST_PORT
Vault38200VAULT_HOST_PORT
Temporal (gRPC)37233TEMPORAL_HOST_PORT
Temporal UI38233TEMPORAL_UI_HOST_PORT
Firecrawl33002FIRECRAWL_HOST_PORT
MinIO (API / console)39000 / 39001MINIO_HOST_PORT / MINIO_CONSOLE_HOST_PORT
Traefik (HTTP / HTTPS)8044 / 8444- (edit Compose)

For day-to-day use you only need 33000 (dashboard) and 38501 (Streamlit chat).

Common commands

CommandBehavior
make bootBootstrap-only (default - postgres, vault, admin-dashboard).
make upAlias for make boot (backward compatibility).
make up-allStart every service at once (CI / debug). Not for production.
make psList running services.
make logsTail logs of the active services.
make downStop all containers (volumes preserved).
make restartmake down + make boot.
make profilesPrint the profile list derived from the manifest.
make helpShow all targets and their descriptions.

Note on make up-all: it starts all services at once (for CI tests and full-stack debugging). Do not use it on the first boot - launching 12 services in parallel before credentials are entered produces "started in the wrong order" errors. Follow the make boot + Setup Wizard flow instead.

Directory layout

PathContents
infra/Docker Compose files (docker-compose.yml, docker-compose.dev.yml) and Postgres migrations.
services/HTTP services - automation-service, assistant-service, admin-dashboard-api, atlassian-mcp (gateway).
workers/Temporal workers - automation-worker, agent-runner-worker, execution-runner-worker.
ui/Front ends - admin-dashboard (Next.js), streamlit-app (Streamlit).
libs/Shared Python libraries (vault_client, audit, llm_client, etc.).
config/Manifest files - services.manifest.json, departments.json and their schemas.
prompts/LLM prompts - task_creation_assistant, assistant_chat, notification templates.
docs/Runbooks, user guide, env reference.
tests/Cross-service property/integration tests (each service also has its own tests/).
scripts/up.sh, up.ps1 wrappers + maintenance scripts.

Architecture

The stack is a set of small services around a shared MCP gateway:

  • Atlassian MCP gateway (services/atlassian_mcp_bitbucket/) - a stateless HTTP surface that exposes Jira, Confluence and Bitbucket tools. Credentials arrive per-request as X-Atlassian-* headers.
  • task-intake-service - receives Jira webhooks and starts workflows.
  • automation-service - governance APIs (departments, credentials, PR review, branch scans) used by the admin dashboard.
  • assistant-service - chat/assist APIs backed by the MCP gateway and an LLM.
  • admin-dashboard (Next.js + API) - the operator/governance UI.
  • streamlit-app - the end-user UI (credentials, chat, task creation).
  • Temporal workers (workers/) - durable execution for long-running automation and runner/SSH steps.

A typical automation flow:

  1. A Jira issue is assigned to the bot - Jira fires a webhook to task-intake-service.
  2. A Temporal workflow starts and asks the LLM what the task needs (code change, docs, a runner, etc.).
  3. The workflow drives the MCP gateway and, when needed, a remote runner over SSH to make changes.
  4. It opens a draft pull request on Bitbucket, publishes a Confluence page, and comments back on the Jira issue.

Built on

  • Atlassian MCP gateway - services/atlassian_mcp_bitbucket/ builds the jellythomas/mcp-atlassian-with-bitbucket fork at a pinned commit. That fork extends sooperset/mcp-atlassian (the original Jira/Confluence MCP server) with Bitbucket support, so all three Atlassian products share one stateless HTTP MCP surface.
  • Temporal - durable workflow execution for the automation tier.
  • HashiCorp Vault - secret storage for department credentials and tokens.
  • Firecrawl - web content retrieval used by research/automation steps.
  • Next.js (admin dashboard) and Streamlit (chat front end) for the two UIs; MinIO for artifact storage and Traefik as the optional edge router.

Next steps

License

This repository is for private use. The bundled Atlassian MCP gateway is built from an upstream fork and retains its original upstream license.

Ver o original no GitHub

Bom para

Relacionados