Skip to content
Free and open source

Your Atlassian work, within reach.

A delivery system should be easier to operate than the work it tracks. I built keen to put Atlassian Cloud within reach of people, scripts and AI agents through one shared set of commands.

REST endpoints, across nine Atlassian products
1,843
Tools an AI agent needs: search, then run
2
Teamwork Graph queries
18
keen finds the command for a plain-English request, then shows the exact request it would send. Recorded against a placeholder site, so no real data is on screen.

The problem

Automating an ordinary delivery task can mean finding an endpoint, rebuilding a request body, looking up an identifier and trimming a large response before the useful work begins. Repeat that across issues, sprints, documentation and administration, and every automation grows its own fragile collection of requests.

AI agents add a second constraint. Putting thousands of operations in front of an agent at once uses up the room it needs to think about the task, and asking it to search live work again and again creates requests nobody needed. The problem was coverage and access together: a broad toolset that stays practical to use.

Who it's for

Delivery teams, platform engineers and product teams who connect Jira and Confluence to automation or AI agents. I am one of them: every ticket, sprint and comment in my own delivery system goes through keen.

What I built

A free, open-source command line tool and MCP server covering 1,843 Atlassian REST endpoints. One command set reaches Jira, Confluence, Agile, Marketplace, Service Desk, Assets, Compass, Bitbucket and Admin, and eighteen Teamwork Graph queries extend it to the relationships between work and teams.

People get discoverable commands and readable output. Automation gets structured responses, non-interactive operation, field selection and a dry run. A local sync gives teams a searchable copy of their work that can still be read when the service cannot be reached. There is no keen account and no telemetry. It installs from Homebrew and is listed in the MCP registry.

How it works

The decision that shaped the product was to give an AI agent two tools instead of 1,843. keen_search finds the operations that match a task and returns their parameters; keen_run runs the one chosen. Finding a command happens locally and never calls Atlassian, so the agent gets a small front door to a large working surface.

Repeat reads within five minutes can come from a cache, and a successful write clears it so the next read sees the change. Coverage is generated from Atlassian's official API specifications, and each refresh is compared with a recorded baseline so my own improvements survive and retired operations are flagged.

What was hard

Keeping broad coverage useful as upstream APIs change. A command can still appear in an official specification after its endpoint stops working, so keen marks deprecated operations, ranks their replacements first and explains the replacement when an old command is gone.

Early use also exposed defects in local search and database migration. Fixing them pushed the product toward everyday tasks rather than a catalogue of endpoints. The count measures coverage; finding the right task, clear errors and fresh reads decide whether that coverage helps someone finish a job.

Get in touch

I am open to Senior Technical Product Manager roles. If you want to talk about how I took this product from problem to launch, write to me.