Is a Jira MCP server and your own skills enough for a team?
Updated
For one developer, yes. A tracker's MCP server, a rules file and a few skills in Claude Code make a working setup in a weekend. This page gives that setup and shows where it stops once three or more people share the tasks.
An MCP server is a program that gives an AI app tools and data over the open Model Context Protocol. A skill is a SKILL.md file of instructions that Claude Code loads when it is relevant, or that you run as /skill-name. A plugin packages skills, agents, hooks and MCP servers so that Claude Code installs them as one unit. A tracker CLI is the vendor's command-line tool, such as Atlassian's acli or GitHub's gh, which the agent runs in the shell.
Jira MCP, Jira CLI or a skill
| Tool | What it gives the agent | How you set it up | A limit the vendor states |
|---|---|---|---|
| Atlassian Rovo MCP server | Tools for Jira and other Atlassian Cloud apps, within the signed-in person's permissions | claude mcp add --transport http atlassian https://mcp.atlassian.com/v2/mcp, then /mcp to sign in with OAuth 2.1 | It needs an Atlassian Cloud site, and some tools consume Rovo credits |
Atlassian CLI (acli) | Commands to search, create, edit and transition work items, which the agent runs in the shell | brew tap atlassian/homebrew-acli, brew install acli, then acli jira auth login --web | Atlassian Government Cloud is not supported, and each version is supported for 6 months |
| A Claude Code skill | Instructions the agent follows; it reaches Jira through an MCP server or a CLI | A .claude/skills/<name>/SKILL.md file committed to the repository | Anthropic advises keeping SKILL.md under 500 lines |
Checked on 13 November 2026.
The setup you can build in a weekend
- Connect a tracker MCP server. Atlassian publishes one for Jira Cloud at
https://mcp.atlassian.com/v2/mcp, and Linear publishes its own athttps://mcp.linear.app/mcp. Both sign you in with OAuth 2.1. For Linear runclaude mcp add --transport http linear-server https://mcp.linear.app/mcp, then/mcpin Claude Code. - Keep a project wiki in the repository. Create
docs/wikiwith pages on concepts, features, decisions and conventions, plus anindex.md, as in the "LLM wiki" approach. After each change the agent updates the wiki from the diff. Our guide is How to set up a project wiki for AI agents. - Write a rules file. In the root
CLAUDE.md, tell the agent to read the wiki index before each task. Claude Code loads it into every session. - Turn repeated prompts into skills. Each one becomes a
SKILL.mdfile in its own folder,.claude/skills/<name>/SKILL.md, committed so the rest of the team gets it too.
For one developer this is enough.
Checked on 13 November 2026.
Where it stops for a team of three or more
The examples below use a made-up team: Ryan and Megan write code, and Emily is the PM.
- Nobody owns a status. A task sits in Review. Ryan thinks Megan starts the review, and Megan thinks the author does.
- Each developer has their own prompts. We retyped "use skill X for task Y" on every task, and on a team each person types it in their own words.
- Answers stay in chat. Emily answers the agent's question in Slack. A week later Megan reads the task and cannot tell Emily's words from the agent's guess.
- Nobody sees which session waits for whom. Ryan's session has waited all morning for Emily's answer, and Emily does not know.
- The PM and QA do not work in a terminal. Skills and
CLAUDE.mdlive in the repository, so Emily asks a developer to run her step.
What a workflow layer adds
CRUX starts from what your tracker already holds. The agent reads the task, its parent and the decisions already on it, and each status in your workflow has its own instructions, so nobody retypes the prompt. CRUX connects as one more MCP server at https://crux.team/mcp/p/<projectId>, with a personal token in the CRUX_MCP_TOKEN environment variable. You set it up in the dashboard dialog "Connect your coding agent", which for Claude Code also adds permissions.allow: ["mcp__crux__*"] to .claude/settings.local.json.
| Team need | What you build yourself | What CRUX gives |
|---|---|---|
| Who acts on a status | A wiki page that names the owners | Each status in your workflow has an owner role, and a person in that role starts the step. |
| One process for everyone | Shared skills and a review of every edit | One workflow for the project, and the agent gets each step's instructions from CRUX. |
| Answers on the task | A habit of pasting answers into comments | The agent asks the person whose role should know. Each step leaves an artifact on the task: its result, which the next step starts from. Text an agent wrote starts with a robot emoji. |
| Sessions that wait | Messages in the team chat | The dashboard home lists the sessions that need you under "Needs you". |
| PM and QA | A coding agent on their machines | An agent that runs on CRUX in the browser, on the project's model key, with the repository read-only. |
The CRUX MCP server has no wiki tools, so an agent on your machine reads the wiki from the repository checkout. Nobody acts under a bot account.
Limits
CRUX is in beta, one founder builds it, and there are no customer stories yet. It is available only as a hosted service. It supports Jira Cloud, Linear, Plane and GitHub Issues as a tracker, and not Jira Server, Jira Data Center or GitHub Enterprise Server.
A Jira admin may have to approve the CRUX app when "Block user apps" is switched on. The agent can move a Jira task only through a direct transition that is available to the person.
One person uses CRUX free. A team of 2 to 10 people pays $49 per active project a month, and 11 or more pay $120. The first month of each project is free. During the beta billing is not switched on, so nothing is charged. The details are on the pricing page.
When doing it yourself is the better choice
- You work alone, or two of you talk every day.
- Only developers touch the tasks.
- Your tracker is not Jira Cloud, Linear, Plane or GitHub Issues.
- Your tasks must not pass through a hosted service.
Next step
Start free, or begin with How to set up a project wiki for AI agents. Other comparisons: spec-driven frameworks and task-to-PR agents.
FAQ
What is the best Jira MCP server?
For Jira Cloud, Atlassian's own Rovo MCP server. Atlassian has no official server for Jira Server or Data Center yet (JRASERVER-78874), so teams there use a community server such as sooperset/mcp-atlassian, the most-starred one on GitHub on 13 November 2026.
Can I build this myself with CLAUDE.md and skills?
Yes, and for one developer we recommend it. The four steps above are the whole recipe. CRUX is for when several people share the tasks and each status needs an owner.
Do I still need the tracker's MCP server with CRUX?
No. The agent reads and updates tasks through the CRUX MCP server. You can keep the tracker's own server for other work.