Connect Claude Code to Jira and Confluence with one claude mcp add command. See the v2 endpoint, scope choices, OAuth sign-in, API token headers, first prompts, a team .mcp.json, and a troubleshooting table for the errors you will actually hit.
Your Jira backlog never shrinks, and nobody can find anything in Confluence. Claude can read both, write to both, and answer questions about them without you pasting a single ticket into a chat window. The whole link is one terminal line: claude mcp add --transport http atlassian https://mcp.atlassian.com/v2/mcp. Run it, sign in once in your browser, and Claude Code can search Jira with JQL, create work items, and read or edit Confluence pages, all under your own permissions.
This walkthrough goes through claude mcp add atlassian in the order you will actually need it: what the server exposes, what to check first, the exact command, the three scopes, OAuth sign-in, a first round of prompts, the API token option for automation, team sharing, and a troubleshooting table. Every command and endpoint below comes from the current Atlassian and Claude Code documentation, checked in October 2026.
MCP (Model Context Protocol) is the open standard that lets an AI client call outside tools. Atlassian hosts its own remote server, the Atlassian Rovo MCP Server, so you install nothing locally. Claude Code connects to a URL, and Atlassian handles authentication, permissions, and the Jira and Confluence calls behind it.
One endpoint serves Jira, Jira Service Management, Confluence, Bitbucket, Projects, Goals, and Loom recordings. This article sticks to Jira and Confluence, the pair most teams connect first. Sign-in uses OAuth 2.1, and every action respects your existing access controls. Claude sees exactly what your account sees, nothing more.
You never type tool names. You write a plain request, and Claude picks the matching tool. Still, knowing the menu helps you phrase better prompts:
| Product | Read | Write | Search |
|---|---|---|---|
| Jira | getJiraIssue, listJiraProjects, listJiraBoards, listJiraIssueComments, listJiraIssueTransitions | createJiraIssue, editJiraIssue, transitionJiraIssue, addOrEditJiraIssueComment | searchJiraIssuesUsingJql |
| Confluence | getConfluenceContent, listConfluenceSpaces, listConfluenceContent, listConfluenceComments | createConfluenceContent, updateConfluenceContent, createConfluenceComment, addLabelsToConfluenceContent | searchConfluence (CQL) |
💡 Tip: Read-only tools are safe to try first. Write tools change real tickets and real pages, so save those for after the connection works.
The checklist is short, but skipping one item is how most setups stall:
claude --version in a terminal. If it prints a version, you are set.
The whole setup takes a few minutes once these boxes are ticked.
Open any terminal and run:
claude mcp add --transport http atlassian https://mcp.atlassian.com/v2/mcp
Each piece does one job:
claude mcp add registers a new MCP server with Claude Code.--transport http selects the streamable HTTP transport that remote servers use.atlassian is the local name. It is what you will see in /mcp and in claude mcp list.https://mcp.atlassian.com/v2/mcp is Atlassian's v2 endpoint.💡 Tip: Older write-ups show a v1 address. Atlassian's current documentation points to v2, so copy the URL above rather than an old forum answer.
By default the command saves the server in local scope. You can choose where it lives with --scope:
| Scope | Flag | Loads in | Shared with team | Stored in |
|---|---|---|---|---|
| Local (default) | --scope local | Current project only | No | ~/.claude.json |
| Project | --scope project | Current project only | Yes, via version control | .mcp.json in the project root |
| User | --scope user | All your projects | No | ~/.claude.json |
For personal use across every repository, user scope is the practical pick:
claude mcp add --transport http atlassian --scope user https://mcp.atlassian.com/v2/mcp
Project scope also works here, because the URL carries no secret. Each teammate still signs in with their own Atlassian account.
Adding the server does not log you in. Finish with these steps:
/mcp.claude mcp list.Prefer to stay in the terminal? Run claude mcp login atlassian. On a headless box, claude mcp login atlassian --no-browser prints a URL instead of opening a browser. To see details for one server, use claude mcp get atlassian. The status column tells you where you stand:
| Status | Meaning |
|---|---|
| ✔ Connected | Server is active and ready |
| ! Needs authentication | OAuth sign-in still pending |
| ✘ Failed to connect | Connection error, usually a URL or network problem |
Start with reads. A good first test proves the sign-in works and shows Claude how your instance is organized.
Ask in plain English, or paste JQL directly:
"List unresolved issues in project PAY updated this week, ordered by priority, and summarize the top five."
Claude calls searchJiraIssuesUsingJql. If you want full control, hand it the query yourself:
project = PAY AND status != Done AND assignee = currentUser() ORDER BY priority DESC
When a name is ambiguous, ask Claude to run lookupJiraAccountId for a teammate before filtering by assignee. It resolves a display name or email to the account ID that JQL needs.
Writes work the same way, with one habit worth adopting: ask for a draft first.
"Draft a Bug for project PAY titled 'Checkout fails on saved cards' using the stack trace above. Show me the fields. Create nothing until I say go."
Behind the scenes Claude looks up the project's issue types and required fields, then calls createJiraIssue once you approve. Moving work along is just as short:
"Move PAY-142 to In Review and add a comment linking the pull request."
That request uses transitionJiraIssue and addOrEditJiraIssueComment. By default Claude Code asks for approval before running MCP tools, so read each write request before you accept it.
💡 Tip: Name the project, issue type, and priority in your first message. Prompts that carry those details need fewer tool calls and fewer follow-up questions, which also keeps Rovo credit use down.
Confluence works through CQL search and page tools:
"Search Confluence for our incident response runbook and summarize the escalation steps."
"Create a page in the ENG space called 'Release checklist' from the notes in this file."
"Add a footer comment to the onboarding page asking whether the VPN steps are still right."
Updates to existing pages can replace the full body or apply granular edits. For anything people rely on, tell Claude to make targeted edits and show the change before saving.
OAuth suits daily interactive work. API token authentication suits pipelines, bots, and other non-interactive runs where no one is around to click a consent screen. It is optional, and an organization admin must enable it in the Rovo MCP server settings first.
Atlassian accepts two header formats:
Authorization: Basic <base64(email:api_token)>Authorization: Bearer <credential>Generate the Base64 value for a personal token on macOS or Linux:
printf '%s' 'you@company.com:YOUR_API_TOKEN' | base64
On Linux add -w 0 so the long output stays on one line. On Windows PowerShell:
[Convert]::ToBase64String([Text.Encoding]::UTF8.GetBytes("you@company.com:YOUR_API_TOKEN"))
Then register the server with a different name, so it can sit next to your OAuth entry:
claude mcp add --transport http atlassian-ci https://mcp.atlassian.com/v2/mcp \
--header "Authorization: Basic BASE64_VALUE"
⚠️ Warning: Treat the token like a password. Keep it in local or user scope, never in a committed
.mcp.json, and rotate it when someone leaves the team.
Atlassian notes that some MCP tools may not be available with token authentication, because tokens carry selectable scopes that are narrower than a full OAuth session. Here is the trade in one table:
| OAuth 2.1 | API token | |
|---|---|---|
| Setup | Browser sign-in through /mcp | Admin toggle plus a header |
| Best for | Daily interactive work | CI jobs, bots, automation |
| Identity | Your user | Your user, or a service account |
| Tool availability | Full set | Some tools may be missing |
To give a whole repository the same connection, add the server in project scope or write the file by hand:
{
"mcpServers": {
"atlassian": {
"type": "http",
"url": "https://mcp.atlassian.com/v2/mcp"
}
}
}
Save it as .mcp.json in the project root and commit it. In interactive sessions Claude Code asks each teammate to approve project servers before using them. In non-interactive modes such as claude -p, project servers load without prompting. To reset those approvals, run claude mcp reset-project-choices.
Atlassian's own notice says: "On March 1, 2027 any existing utilization of v1 will automatically start to expose and utilize v2 tools." Clients that cached old credentials may need to clear their cached clientIds or .well-known credentials. If you registered a v1 address months ago, do a clean swap now:
claude mcp logout atlassian
claude mcp remove atlassian
claude mcp add --transport http atlassian https://mcp.atlassian.com/v2/mcp
Then run /mcp and sign in again.
Two habits keep the setup healthy over time:
Most failures fall into a handful of patterns. Start with claude mcp list, then match the symptom:
| Symptom | Likely cause | Fix |
|---|---|---|
! Needs authentication | Sign-in never finished or has expired | Run /mcp or claude mcp login atlassian |
✘ Failed to connect | Wrong URL, wrong transport, or a network block | Run claude mcp get atlassian and confirm the type is http and the URL ends in /v2/mcp |
| Browser never opens | Remote or headless session | Use claude mcp login atlassian --no-browser |
| Sign-in loops after switching versions | Stale cached client data | Run the logout, remove, and add sequence above |
| An expected tool is missing | Token scopes narrower than OAuth | Widen the token scopes or switch to OAuth |
| Searches return nothing for a project you can open | Name mismatch | Ask Claude to list the Jira projects it can see, then reuse the exact name |
When nothing else works, removing and re-adding the server takes about ten seconds and rules out a bad local entry. After any change, rerun claude mcp list before blaming your prompt. A green ✔ Connected line tells you the problem sits in the request, not the connection.
Connecting Claude to Atlassian solves the plumbing. Two jobs still sit around it: writing ticket text worth reading, and giving Confluence pages some visual weight. PicassoIA helps with both, as a separate tool next to your Claude Code setup. The MCP link above runs in Claude Code, while the steps below happen on PicassoIA.
Claude Sonnet 5 turns rough notes into clean ticket drafts before you ever open a terminal:
low skips deep thinking and answers fastest. Move to high or max when the problem touches several systems.Need heavier reasoning for a tangled incident write-up? Claude Opus 4.7 and Claude Fable 5 are on the same platform.
A runbook with a clear header image gets opened more often than a wall of text. P-Image, Flux 2 Pro, Seedream 5 Pro, and GPT Image 2 all turn a sentence into a photo you can use as a page banner. Try a prompt such as "a sunlit team whiteboard with sticky notes, shallow depth of field, natural film grain."
The Atlassian tool list includes attachment upload for Jira issues but only downloads for Confluence, so drop the finished image into the page with the Confluence editor.
PicassoIA also offers its own MCP connection for image and video generation, managed from your account. The idea is the same as the one you just set up: connect once, then ask in plain English.
Run the command, sign in, and ask Claude for your first JQL search. Then open PicassoIA, pick a model, and create the ticket drafts and images your team keeps postponing. Your next sprint review will thank you.
