jira-mcp

MCP server for Jira integration -- multi-instance routing, ADF formatting, task caching, and comment templates via the Model Context Protocol.

What's New in v1.8.1

• Duplicate detection actually works now. The lookup used summary = "...", and Jira answers = on a text field with an empty result set rather than an error -- so it always reported "not found". update_existing never updated anything, and re-running a bulk config duplicated every task. Now a quoted ~ phrase with an exact comparison on the result.

• status takes a path, not just a name. "status": ["On hold", "Open"] walks the transitions in order. Jira only ever exposes the transitions out of an issue's current status, so a target that is not directly reachable from the initial status was previously impossible to set -- which is exactly how workflows that gate Open behind a Reopen transition behave.

• Failed transitions stop being silent. The old code caught every transition error and moved on: issue created, status ignored, run reports failed: 0. It now returns a warning naming the status that failed and listing what was reachable instead, surfaced per task and aggregated into create_monthly_tasks output.

• TaskResult.warning -- new string | null field for problems that occur after the issue is already written. formatBulkResult prints it under the task line.

See CHANGELOG.md for full details.

Related MCP server: Jira MCP Server

Table of Contents

• Install

• Configuration

• CLI Commands

• Available Tools

• Comment Templates

• Usage with Claude Code

• Usage with Other MCP Clients

• Architecture

• Key Features

• Documentation

• Contributing

• Security

• License

Install

npm install -g @softspark/jira-mcp
# or use directly
npx @softspark/jira-mcp

Update

npm update -g @softspark/jira-mcp

Configuration

1. Generate an API token

Go to Atlassian API Tokens and create a token.

2. Initialize global config

jira-mcp config init
jira-mcp config set-credentials
jira-mcp config add-project

All configuration lives in ~/.softspark/jira-mcp/ (created by jira-mcp config init). This is the standard config directory for all SoftSpark open-source tools.

CLI Commands

Available Tools

Comment Templates

8 built-in templates organized by category:

Using Templates

Example with Claude Code:

"Add a status update to PROJ-123: completed auth module, next steps are testing, no blockers"

The AI will use add_templated_comment with template_id: "status-update" automatically.

File-Backed Overrides

System templates are shipped as markdown files. User overrides are loaded from:

~/.softspark/jira-mcp/templates/comments/
~/.softspark/jira-mcp/templates/task-templates/

If a user file has the same id as a system template, the user file wins globally for all projects.

jira-mcp template add comment ./my-status-update.md
jira-mcp template add task ./my-bug-task.md
jira-mcp template list

Usage with Claude Code

Add to your Claude Code MCP configuration (~/.claude/claude_desktop_config.json or project-level):

{
  "mcpServers": {
    "jira": {
      "command": "npx",
      "args": ["@softspark/jira-mcp"]
    }
  }
}

Configuration is automatically loaded from ~/.softspark/jira-mcp/. Run jira-mcp config init first to set up.

AI Toolkit Rules

If you use @softspark/ai-toolkit, register the Jira rules so that Claude Code automatically follows project conventions (language checks, sync-first workflow, time format, etc.):

ai-toolkit rules add jira-mcp --path /path/to/jira-mcp/rules/jira-mcp.md

Or copy rules/jira-mcp.md to your ai-toolkit rules directory manually. The rules file covers:

• Language first -- always check project language before writing comments/descriptions

• Sync before read -- cache may be stale

• Status transitions -- check valid transitions before changing status

• Time format -- "2h 30m", never days

• All 19 MCP tools and 20 CLI commands reference

AI Toolkit Hooks

To require explicit user approval before Jira comment writes, inject the repo-owned hook manifest:

ai-toolkit inject-hook https://raw.githubusercontent.com/softspark/jira-mcp/main/hooks/jira-mcp-hooks.json

This installs a PreToolUse guard for add_task_comment and add_templated_comment. The hook blocks the tool call, shows the exact comment preview, and tells the agent to retry only after the user approves it with user_approved=true.

The MCP server still enforces user_approved=true at runtime, so the hook is UX guidance plus an extra safety layer rather than the only check.

Usage with Other MCP Clients

Any MCP client that supports stdio transport can use this server:

{
  "command": "npx",
  "args": ["@softspark/jira-mcp"],
  "transport": "stdio"
}

Configuration is loaded from ~/.softspark/jira-mcp/ automatically. No environment variables needed.

Architecture

src/
  adf/            Atlassian Document Format conversion (MD <-> ADF)
  bulk/           Bulk task creator (dry-run, rate limiting, bilingual)
  cache/          Local task, workflow, and user caching (JSON files)
  cli/            CLI entry point and subcommand handlers
    commands/
      cache/      Cache management subcommands
      config/     Config management subcommands
      template/   File-backed template management subcommands
      create.ts   Bulk task creation command
      create-monthly.ts  Monthly admin task automation
  config/         Configuration loading and Zod validation
  connector/      Jira API client (built-in fetch, instance pool)
  errors/         Typed error hierarchy
  operations/     Business logic (status, comments, time tracking)
  templates/      File-backed comment/task template loading and registries
  tools/          MCP tool handlers (19 tools, one file per tool)
  types/          Shared TypeScript types
  server.ts       MCP server setup and tool registration
  cli.ts          CLI entry point

Key Features

Multi-instance routing -- single server manages multiple Jira Cloud/Server instances. Project key determines which instance handles the request. Connectors deduplicated by URL via InstancePool.

ADF round-trip -- comments and descriptions use Atlassian Document Format natively. Markdown in, ADF to Jira v3 API, ADF back to readable markdown. Never loses formatting. See ADF Reference.

Local caching -- tasks synced to ~/.softspark/jira-mcp/cache/ with atomic writes (tmp + rename). Work offline with read_cached_tasks, sync on demand. Workflow and user caches for status validation and assignee resolution.

File-backed templates -- 8 built-in comment templates and built-in task templates are stored as markdown files, with global user overrides loaded from ~/.softspark/jira-mcp/templates/. Both support {{variable}} interpolation and {{#var}}...{{/var}} conditional blocks. See Templates Reference.

Language configuration -- global default_language with per-project override. Supports: pl, en, de, es, fr, pt, it, nl. get_project_language tool and language field in get_task_details let AI assistants write content in the correct language. See Configuration.

Bulk task creation -- create tasks from JSON configs with dry-run default, rate limiting, epic link discovery, assignee caching, multilingual support (8 languages), and idempotent updates. See CLI Usage.

Per-instance credentials -- different API tokens per Jira instance URL. Single-credential format still works (backward compatible). See Configuration.

Supply chain protection -- ignore-scripts=true, no axios, no dynamic requires. Self-contained 535KB bundle, 1 runtime dep (commander).

Typed error hierarchy -- 20 error classes with machine-readable codes. Every tool returns structured { success, error, code } responses. No stack traces leak to MCP clients.

Strict TypeScript -- strict: true, no any, readonly interfaces, Zod validation at all boundaries, 680 tests across 65 test files. Self-contained 535KB package.

Documentation

Contributing

See CONTRIBUTING.md.

Security

See SECURITY.md.

License

Apache License 2.0 -- see LICENSE and NOTICE. SoftSpark.

Fork it, modify it, ship it commercially. Three things the licence asks in return:

• Keep the attribution. Redistributions must carry the contents of NOTICE (§4d) -- project name, copyright and source URL.

• Say what you changed. Modified files must carry a prominent notice stating that you changed them (§4b).

• Names are not included. No rights to the "jira-mcp" or "SoftSpark" names or marks (§6).

The published bundles are minified, so the source headers do not survive into dist/. Attribution reaches consumers through a build banner instead -- the first comment in dist/index.js and dist/cli.js.

Releases up to and including v1.6.0 were published under MIT and stay available under MIT; the change applies going forward and revokes nothing already granted.

Changelog

See CHANGELOG.md.

Built at SoftSpark. Designed for AI-assisted Jira workflows.