Notes from the work

One email when a tutorial or migration write-up goes live. Nothing else, and one click to leave.

LeanZero

Two people in Romania doing Atlassian migrations, Forge apps and practical AI work for teams that would rather talk to the person doing the job. Most of what we learn ends up on this site.

Services

  • Atlassian Migrations
  • Atlassian FastShift
  • Atlassian Maintenance
  • Forge App Development
  • AI Development Consultation

Topics

  • Jira
  • Jira Service Management
  • Confluence
  • Bitbucket
  • Atlassian Forge
  • Cloud Migration
  • Local AI
  • AI Coding
  • Certifications
  • Atlassian Team
  • All topics

Company

  • Blog
  • Tutorials
  • Contact

Community

  • Join Discord
  • Support this site

© 2026 LeanZero. All rights reserved.

Privacy PolicyTerms of Service24/7 SupportTrust CenterSecurityDPALegal notice
  1. Home
  2. Ai Tools
AI Tools, 6 we built and use

AI tools for Atlassian

A Jira Forge app, two MCP servers, a local-model agent and a skills pack. All in daily use; the servers, the agent and the skills have public code.

See the 6 toolsAI terms explained
Tools
6
MCP tools
28
Jira events
68
To try
€0

The tools, and what each one is for

01

CogniRunner

Forge app · Jira workflow AI

Product page

AI validators, conditions and post-functions on Jira transitions, written in plain English instead of Groovy. It reads the issue the way a reviewer would: a validator can fail a transition because the description does not match the summary, not only because a field is empty.

What it does

  • AI validators that check meaning, plus zero-AI conditions and post-functions
  • Listeners on 68 Jira events and cron-scheduled jobs
  • Duplicate detection and attachment analysis on the issue itself
  • Rules REST API for managing rules from your own scripts

Plain facts

  • Listed on the Atlassian Marketplace
  • Runs on Forge, inside Atlassian's cloud
  • 22 video tutorials on the product page
  • Proprietary, paid via Atlassian
Product page
02

MCP Web Search

MCP server · live web for agents

Get a demo key

Gives any MCP-capable AI client a live view of the web: search, page and PDF extraction, GitHub and OpenAPI reading. We wrote it because Claude Code kept answering Forge questions from a manifest format that no longer existed.

What it does

  • 11 tools: search, fetch, extract, read GitHub repos and OpenAPI specs
  • Cited results, so the agent can show where a claim came from
  • Self-host in about five minutes, or run it as an always-on HTTPS service

Plain facts

  • Free demo key, no account, no card
  • Node.js and TypeScript
  • Source on GitHub
Get a demo key
03

MCP Document Processor

MCP server · office files for agents

Get a demo key

Lets an agent read, create and edit PDF, DOCX, Excel and PowerPoint files. Parsing happens locally with no network access, so the only thing that reaches a model is what your client chooses to send it.

What it does

  • 17 tools across PDF, DOCX, XLSX and PPTX
  • OCR for scanned, image-only PDFs
  • Format-preserving edits instead of regenerating the whole file
  • Jira integration for attachments

Plain facts

  • Free demo key, no account, no card
  • Files are parsed on the server you run it on
  • Source on GitHub
Get a demo key
04

LeanZero Link

Part of Goose Swarm · private mesh

Product page

A feature of Goose Swarm, available only inside the Goose Swarm app for now. It joins your Macs running Goose Swarm into one private mesh, so one Mac can see the others and send an agent prompt to another. You sign in with your email and a 6-digit code; there is no key to copy.

What it does

  • Device list, availability, sessions and a live stream from your other Macs
  • Remote runs stay off until the target Mac's owner switches them on
  • Runs its own tailscaled, separate from your own Tailscale

Plain facts

  • Part of Goose Swarm only, for now (since 2.0.3)
  • One isolated mesh per account
  • Sign-in service can be self-hosted
Product page
05

goose local edition

Agent · local models only

Product page

Our fork of the open-source goose agent. It decomposes a spec into a task graph, runs the tasks across a fleet of small local models on LM Studio, then verifies its own work by actually running it. Nothing leaves the LAN.

What it does

  • Spec to task graph to working build, with a repair loop when a step fails
  • Swarms several local models instead of relying on one large one
  • Scored publicly on our agentic benchmarks

Plain facts

  • Rust, fork of block/goose
  • Runs on LM Studio, fully offline
  • Source on GitHub
Product page
06

Agent skills for Atlassian

Skills pack · for coding agents

Browse the skills

Instruction sets we extracted from shipping Forge apps and migration scripts, so a coding agent stops inventing manifest keys, breaking ADF trees and 400-ing on JQL. They are what our own agents run with.

What it does

  • 9 skills: Jira Forge, Confluence Forge, org admin, Jira API, Confluence API, migration scripts, Cloud-to-Cloud migration, Forge security review, Automation rules
  • Each one states when to use it and when to skip it
  • Works with Claude Code and any agent that reads SKILL.md files

Plain facts

  • Open source
  • Extracted from production apps
  • Updated as the apps change
Browse the skills

Want one of these in your own Jira?

A Forge prototype starts at €500, priced before we start. Or ask on Discord first, that part is free.

Forge app developmentTalk to us

Before you install anything

Which of these run inside Atlassian?open
Only CogniRunner. It is a Forge app, so it runs in Atlassian's own cloud and appears in the Jira workflow editor like a native validator. The two MCP servers and goose run wherever you run them: your laptop, a VM, or the hosted demo we keep up for the free keys.
Does my data go to an AI vendor?open
The MCP servers do not call a model themselves; your AI client does. The document processor parses PDF, DOCX, XLSX and PPTX locally with no network access, so only what your client sends to its model leaves the machine. CogniRunner uses the provider key you configure, so you decide where prompts go. goose is local models only.
What does it cost to try?open
The MCP servers have a free demo key with no account and no card. goose and the skills are open source. CogniRunner is on the Atlassian Marketplace and the listing carries the pricing.
Can you wire one of these into our Jira?open
Yes, that is the Forge app development service. A prototype starts at €500, UAT is priced after you have seen it, and production work is billed at €50 an hour with no retainer.
Are they maintained?open
They are the tools I use on client work every week, so they get fixed when they break on me. Every one except CogniRunner, which is proprietary, has a public GitHub repository, and the commit history is the honest status page.

If you want the background first

AI Glossary

Plain-English definitions of the terms these pages use: MCP, RAG, quantisation, agent.

Tutorials

Step-by-step write-ups, with the code, of the setups above and the ones that failed.

AI for your team

The readiness sprint: local or private AI set up for a dev team, then handed over. From €1,200.