Acendas Atlassian Suite

Comprehensive Claude Code plugin for Jira Cloud + Confluence Cloud + Bitbucket Cloud, built on a single Node MCP server (@acendas/atlassian-mcp) with 29 workflow skills and an orchestrator agent.

What's in the box

Component	Count
MCP server (Node, FastMCP)	1
MCP tools	179
Skills	29
Agents	7 (1 router + 6 specialists)

MCP tool coverage

Product	Tools	Coverage
Bitbucket Cloud	122	PRs, branches, branch restrictions, repos, file contents, forks, commits, comments, tags, pipelines, schedules, pipeline variables (repo/workspace/project), deployments, environments, env variables, code insights (reports + annotations + build statuses), native issues, snippets, webhooks, projects, code search, deploy keys, SSH keys, users, workspace metadata + members + permissions
Jira Cloud	38	Issues (CRUD + batch), transitions, comments, worklogs, watchers, links, remote links, changelogs, fields, custom fields, projects, components, versions, users, agile boards, sprints, JQL search
Confluence Cloud	19	Pages (CRUD), search (CQL), space tree, version history, version diff, comments + replies, labels, attachments, spaces, user search

Skills (29)

Cross-product (11): init, link-pr-to-issue, pr-summary-to-jira, sprint-status, standup-notes, sprint-retro, release-notes, publish-release-notes, review-pr, create-issue, triage-issue

Bitbucket (8): pr-list, pr-create, pipeline-status, pipeline-trigger, deployments, code-search, commit-show, branch-protection

Jira (5): jql-search, jira-issue, jira-sprint, jira-link, jira-worklog

Confluence (6): confluence-search, confluence-page, confluence-edit, confluence-tree, confluence-comment, confluence-attachment

All user-invokable as /atlassian-suite:<skill-name>.

Agents (7)

Each agent has a focused tool surface + specialized system prompt for autonomous, multi-step work.

Agent	Owns	Color
atlassian-orchestrator	Router — dispatches to specialists or coordinates multi-specialist flows	blue
code-review-orchestrator	Bitbucket PRs — review, comments, approve/decline, reviewers, with Jira context	cyan
sprint-orchestrator	Jira Agile — boards, sprints, planning, retros, standup, active-sprint health	green
release-orchestrator	Bitbucket merged PRs + Jira fixVersion + tags + Confluence publish	orange
devops-orchestrator	Pipelines, deployments, environments, branch protection, code insights, schedules, variables	red
triage-orchestrator	Jira issue lifecycle — bulk triage, create, link, watchers, batch transitions	yellow
knowledge-orchestrator	Confluence read/write/edit — pages, spaces, comments, attachments, version diff	purple

For most tasks, dispatch the specialist directly. Use atlassian-orchestrator only when intent is unclear or the work spans 3+ specialists end-to-end.

PR review pipeline (8 agents)

code-review-orchestrator runs a multi-scanner pipeline modeled on Shipyard's pattern. All scanners write strict YAML output with confidence ≥ 80 and a TRUNCATED protocol for large diffs.

Wave 1 — 6 parallel scanners (sonnet, ≤8 files per round, single-responsibility):
  • pr-review-bugs            — logic errors, null handling, races, leaks
  • pr-review-security        — injection, auth bypass, secrets, crypto
  • pr-review-silent-failures — empty catches, swallowed errors, masked failures
  • pr-review-patterns        — project rules, naming, anti-patterns, duplication
  • pr-review-tests           — coverage, weak assertions, missing edge cases
  • pr-review-spec            — Jira AC compliance, under/over-building
       ↓
Wave 2 — pr-investigator (opus): deep-dives on borderline findings (conf 80–89, must-fix on auth/payments)
       ↓
Wave 3 — pr-critic (opus, high-stakes only): anti-sycophancy adversarial pass with assumptions + pre-mortem + structured criteria
       ↓
Consolidated verdict + optional inline-comment posting

Use via /atlassian-suite:review-pr <pr-id> [--high] [--quick] or by dispatching code-review-orchestrator directly for batch reviews.

Prerequisites

• Node 20+ (node --version)
• pnpm (npm install -g pnpm)
• Atlassian Cloud account with API access
• Bitbucket Cloud workspace

Installation

1. Clone or copy this repo into your Claude Code plugins directory.

2. Build the MCP server:

   cd server
   pnpm install
   pnpm build

3. Configure credentials. Two options — pick one:

   Option A — File (recommended). Run /atlassian-suite:init after first launch. The skill runs an interactive wizard: it opens the Atlassian token page in your browser, prints the exact OAuth scopes to tick per product (Jira / Confluence / Bitbucket), collects URL/email/workspace/token via in-chat prompts, persists to ~/.acendas-atlassian/config.json (mode 0600, atomic write with rolling backup), and self-tests each product against its API. Note: tokens entered in the wizard appear in your Claude Code transcript — if that's a concern, use Option B (env vars) instead.

   Option B — Env vars. Set in ~/.zshrc / ~/.bashrc / project .env:

   # Jira + Confluence — shared Atlassian account
   export JIRA_URL="https://acme.atlassian.net"
   export CONFLUENCE_URL="https://acme.atlassian.net/wiki"
   export ATLASSIAN_USERNAME="you@example.com"
   export ATLASSIAN_API_TOKEN="ATATT3xFfGF0..."
   
   # Bitbucket — separate credential (usually a separate account/token)
   export BITBUCKET_WORKSPACE="acme"
   export BITBUCKET_USERNAME="you@example.com"
   export BITBUCKET_API_TOKEN="ATATT3xFfGF0..."   # or app-password / access token

   Env vars take precedence over the file when set. Use them for CI or per-session overrides.

Credential separation

Jira + Confluence share one Atlassian account + one Atlassian API token (from https://id.atlassian.com/manage-profile/security/api-tokens). Bitbucket Cloud is separate — it uses an app password, a Repository / Project / Workspace Access Token, or an Atlassian API token tied to a Bitbucket-enabled account, and usually under a different email/identity than Jira/Confluence.

• For Jira + Confluence, use the shared atlassian.* block (or ATLASSIAN_USERNAME + ATLASSIAN_API_TOKEN).
• For Bitbucket, ALWAYS use the per-product bitbucket.* block (or BITBUCKET_USERNAME + BITBUCKET_API_TOKEN) so an Atlassian-only token never leaks through to Bitbucket calls.

The atlassian.* shared fallback only activates when per-product fields are unset — so if you have a single token that legitimately works for all three, setting just atlassian.* works. If you have separate tokens (the common case), set Bitbucket explicitly.

4. Restart Claude Code. The MCP server starts automatically on first use and reads credentials at startup.

Credential resolution order

For every value (URL, username, token, filter):

1. Per-product env var (JIRA_USERNAME)
2. Shared env var (ATLASSIAN_USERNAME)
3. Per-product file entry (jira.username in config.json)
4. Shared file entry (atlassian.username)
5. Not configured → product disabled

Three credential management tools are always available:

• configure_credentials — write/update the file (token values not echoed in plaintext)
• get_credentials_status — inspect current state, with resolution source per value
• clear_credentials — delete the file (env vars unaffected)

Stack

• Server framework: fastmcp (Node TypeScript port)
• Jira client: jira.js — typed Cloud REST + Agile
• Confluence client: confluence.js — typed Cloud REST v1+v2
• Markdown → ADF: @atlaskit/editor-json-transformer + @atlaskit/editor-markdown-transformer (heavy/accurate)
• ADF → Markdown: adf-to-md (mature standalone)
• Bitbucket: raw fetch against the v2 REST API (no mature typed client; not needed)

All API calls verified against the published .d.ts files of jira.js / confluence.js and Atlassian's official Bitbucket Cloud REST API docs.

Auth

• Cloud only. API Token + Basic Auth across all three products.
• Read-only mode: READ_ONLY_MODE=true blocks writes.
• Filters: JIRA_PROJECTS_FILTER, CONFLUENCE_SPACES_FILTER (comma-separated keys).

Development

cd server
pnpm install
pnpm dev          # tsx — runs server.ts directly
pnpm typecheck    # type-check without emitting
pnpm build        # produce dist/ for production use
pnpm clean        # remove dist/

Plugin layout

atlassian-plugin/
├── .claude-plugin/plugin.json
├── .mcp.json                  # node ${CLAUDE_PLUGIN_ROOT}/server/dist/server.js
├── agents/atlassian-orchestrator.md
├── skills/                    # 29 user-invoked workflows
└── server/                    # Node MCP server
    ├── package.json
    ├── tsconfig.json
    └── src/
        ├── server.ts          # FastMCP entrypoint
        ├── common/            # config, http, ADF, jira/confluence clients
        ├── jira/              # 38 tools across search/issues/projects/agile
        ├── confluence/        # 19 tools across pages/comments/spaces/labels/diff
        └── bitbucket/         # 122 tools across 21 modules