jtk
# jtk — Jira Toolkit
[](https://github.com/zach-snell/jtk/actions/workflows/ci.yml)
[](https://goreportcard.com/report/github.com/zach-snell/jtk)
[](https://zach-snell.github.io/jtk/)
[](LICENSE)
The most comprehensive dedicated Jira MCP server in the open-source ecosystem. A dual-mode Go binary that works as both a rich CLI tool and an MCP server for AI agents.
**11 MCP tools · 4 prompt templates · Full CLI · Single binary · Zero dependencies**
## Why jtk?
| | jtk | mcp-atlassian (Python) |
|---|---|---|
| **Language** | Go (single ~15MB binary) | Python (pip install + deps) |
| **Jira tools** | 11 dedicated tools | ~30 mixed Jira+Confluence |
| **Startup** | ~50ms | ~2s |
| **Permission introspection** | Dynamic at startup | None |
| **Dev status API** | Branches, PRs, commits | Not available |
| **Issue metrics** | Cycle time, lead time, time-in-status | Not available |
| **MCP prompts** | 4 built-in templates | None |
| **Auth** | Classic + scoped tokens | Classic only |
## Features
- **Dual Mode** — CLI for humans, MCP server for AI agents, same binary
- **Git Awareness** — Auto-detects Jira issue keys from branch names (e.g., `feature/PROJ-123-add-login` → `PROJ-123`)
- **Dynamic Permission Introspection** — Queries `/mypermissions` at MCP startup, only registers mutation tools your token allows
- **Dev Status API** — Surfaces branches, PRs, and commits linked to any issue via Jira's 3-step dev-status endpoint
- **Issue Metrics** — Cycle time, lead time, time-in-status breakdown with status transition history
- **MCP Prompts** — Standup summary, sprint status, release notes, dev dependency tree
- **Token-Efficient** — Consolidated action-based tools minimize schema overhead. ResponseFlattener strips bloated JSON. Full ADF↔Markdown conversion
- **Agile-First** — Boards, sprints, backlogs, sprint mutations, active sprint detection
## Installation
```bash
# From source
git clone https://github.com/zach-snell/jtk.git && cd jtk
./install.sh # builds and copies to ~/.local/bin
# Or build manually
go build -o jtk ./cmd/jtk
```
Pre-built binaries available on the [Releases](https://github.com/zach-snell/jtk/releases) page.
## Quick Start
```bash
# Authenticate
jtk auth
# Get current issue from git branch
jtk issues get
# Search with JQL
jtk issues search --jql "project = PROJ AND status = 'In Progress'"
# Create an issue
jtk issues create --project PROJ --type Task --summary "Fix login bug"
# Sprint overview
jtk boards list
jtk boards active-sprint --board 1
```
## CLI Commands
```
jtk auth Authenticate with Jira Cloud
jtk issues Issue CRUD, search, comments, transitions, links
jtk boards Agile boards, sprints, backlogs
jtk projects List, get, create projects
jtk users Search and get users
jtk versions Project versions/releases
jtk worklogs Time tracking
```
## MCP Server
### Stdio Transport (Claude Desktop, Cursor, OpenCode, etc.)
```json
{
"mcpServers": {
"jira": {
"command": "/path/to/jtk",
"args": ["mcp"],
"env": {
"JIRA_DOMAIN": "your-domain",
"JIRA_EMAIL": "you@example.com",
"JIRA_API_TOKEN": "your-api-token"
}
}
}
}
```
### Streamable HTTP Transport
```bash
jtk mcp --port 8080
```
### Environment Variables
| Variable | Description | Required |
|----------|-------------|----------|
| `JIRA_DOMAIN` | Atlassian domain (e.g., `acme` for acme.atlassian.net) | Yes |
| `JIRA_EMAIL` | Email for the API token | Yes |
| `JIRA_API_TOKEN` | Atlassian API token | Yes |
| `JIRA_TOKEN_TYPE` | `classic` or `scoped` (auto-detected if omitted) | No |
| `JIRA_DISABLED_TOOLS` | Comma-separated tool names to hide | No |
## MCP Tools (11)
| Tool | Actions |
|------|---------|
| `manage_issues` | get, create, update, assign, transition, delete, add_comment, edit_comment, list_comments, list_types, get_links, get_history, link, list_link_types, get_watchers, add_watcher, remove_watcher |
| `manage_search` | jql, quick |
| `manage_boards` | list_boards, get_board, list_sprints, get_sprint_issues, get_backlog, get_active_sprint, search_sprints, create_sprint, update_sprint, move_to_sprint |
| `manage_projects` | list, get, list_statuses, create |
| `manage_devinfo` | get_dev_info |
| `manage_worklogs` | list, add |
| `manage_versions` | list, get, create |
| `manage_attachments` | list, download, upload, delete |
| `manage_users` | get_current, search, get |
| `manage_metrics` | get_dates, get_metrics |
### MCP Prompts (4)
| Prompt | Description |
|--------|-------------|
| `standup_summary` | Generate a standup report from recent activity |
| `sprint_status` | Analyze sprint health and progress |
| `release_notes` | Draft release notes from a version's issues |
| `dev_dependency_tree` | Map development dependencies across linked issues |
## Security
**Three-layer safety model:**
1. **Token scopes** — Atlassian scopes control which APIs the token can call (403 if missing)
2. **Permission introspection** — jtk queries `/mypermissions` at startup and dynamically hides mutation tools your account lacks
3. **Tool denial** — Explicitly hide tools: `JIRA_DISABLED_TOOLS="manage_boards,manage_worklogs"`
### Token scopes — use classic
When creating an API token with scopes, **prefer classic scopes** for full read + write:
```
read:jira-user,read:jira-work,write:jira-work,manage:jira-project
```
(Read-only: `read:jira-user,read:jira-work`.) jtk auto-detects the token type, so granular scopes also work — but granular is **Beta and finicky**: "fat" endpoints require every satellite scope (e.g. `/myself` needs `read:user` + `read:avatar` + `read:group` + `read:application-role`), and granular *personal API tokens* currently return `401 Unauthorized; scope does not match` on POST requests through the gateway, making them effectively read-only. `jtk auth` prints copy-paste scope lists for both.
## Development
```bash
go test -race ./... # Run tests
golangci-lint run ./... # Lint
go build -o jtk ./cmd/jtk # Build
```
## License
[Apache 2.0](LICENSE)
TDQS
Scored across 10 tools
Each tool is scoped to a distinct Jira domain (issues, projects, users, boards, worklogs, attachments, etc.), so the boundaries are clear. The only potential overlap is between searching within manage_search and listing within other tools, but the resource-specific naming prevents real ambiguity.
All tools follow a consistent manage_<plural_noun> convention, and each tool exposes lowercase verb-style actions. This creates a predictable pattern that makes it easy to infer what each tool does.
Ten tools is well-scoped for a Jira server. The count is large enough to cover distinct functional areas without becoming unwieldy, and the grouping of many actions under manage_issues keeps the surface organized.
Core Jira workflows are well covered: issues, projects, search, comments, attachments, worklogs, users, boards, sprints, and dev info. Minor gaps exist in version and worklog management (no create/update/delete for versions, no edit/delete for worklogs), but these are edge operations rather than critical missing workflows.