Jira MCP Server: Cloud Only, and What It Can Do
Yes — Atlassian ships an official Jira MCP server, and it went generally available on February 4, 2026. It’s now branded the Atlassian Rovo MCP Server, it’s hosted by Atlassian, and it covers Jira, Confluence, Jira Service Management Ops, Bitbucket Cloud, and Compass through one endpoint.
The single most important thing to know before you spend an afternoon on it: it is Cloud only. If you run Jira Data Center or Jira Server, the official server cannot connect to you at all, and there is no published timeline for that changing. The rest of this covers the exact tools it exposes, how auth and admin approval actually work, the rate-limit problem Atlassian hasn’t addressed, and what people use community servers for instead.
Jira is one of many — the complete list of MCP servers covers every app that ships one, grouped by category.
Jira Cloud only — Data Center and Server are not supported
This is the question that wastes the most time, so it goes first. Atlassian’s own developer docs describe it as “a cloud-hosted Model Context Protocol (MCP) server,” and on the community thread asking directly about Server and Data Center, an Atlassian staffer answered:
“the official Atlassian MCP (Model Context Protocol) Server is currently designed and supported only for Atlassian Cloud products… An active Atlassian Cloud site is a prerequisite for setup and operation.”
That isn’t a gap waiting to be filled. mcp.atlassian.com is a multi-tenant hosted endpoint that authenticates against Atlassian Cloud identity — a self-hosted Data Center instance behind your firewall has nothing on the other end to talk to. No Atlassian doc, blog post, or repo README mentions Data Center support, and none commits to a date.
If you’re on Data Center, skip to the community server section below.
The endpoint, and the one that just died
| Endpoint | Transport | Status |
|---|---|---|
https://mcp.atlassian.com/v1/mcp/authv2 | Streamable HTTP | Current, recommended |
https://mcp.atlassian.com/v1/mcp | Streamable HTTP | Supported |
https://mcp.atlassian.com/v1/sse | HTTP+SSE | Retired June 30, 2026 |
The SSE endpoint is worth calling out because a lot of tutorials still print it. Atlassian deprecated it in March — the connections were “fragile during network drops” — and shut it off on June 30. If you followed a guide and your client won’t connect, check whether it’s pointed at /v1/sse. That’s the most common cause right now.
What the Jira MCP server can actually do
Atlassian publishes the tool list, and it’s worth reading literally rather than trusting a vendor’s summary. Sixteen Jira tools ship today:
Reading: getJiraIssue, searchJiraIssuesUsingJql, getVisibleJiraProjects, getTransitionsForJiraIssue, getJiraIssueRemoteIssueLinks, getJiraIssueTypeMetaWithFields, getJiraProjectIssueTypesMetadata, getIssueLinkTypes, lookupJiraAccountId, atlassianUserInfo, getAccessibleAtlassianResources
Writing: createJiraIssue, editJiraIssue, transitionJiraIssue, addCommentToJiraIssue, addWorklogToJiraIssue
So full JQL search, issue creation, field edits, status transitions, comments, and worklogs. What’s missing is deletion — there is no delete-issue tool, and Confluence likewise has no delete-page tool. Writes are create-and-update only, which is a sensible blast-radius decision on Atlassian’s part.
Confluence gets twelve tools (getConfluencePage, createConfluencePage, updateConfluencePage, searchConfluenceUsingCql, and the comment set) — see Confluence MCP for that side. GA also added semantic search across Jira and Confluence, so “find the ticket about the login timeout” works without you writing the JQL.
Auth, and the admin gate people hit
Auth is OAuth 2.1 with browser-based 3LO consent. The server acts strictly within the signed-in user’s existing Jira permissions — it cannot see a project you can’t see — and Atlassian states it doesn’t store or cache your Jira and Confluence data.
Three things trip people up:
- The first connection installs an app on your site. Atlassian’s docs are explicit that the first user to complete the consent flow must already have access to the products the MCP scopes request. If you’re not that person, you may need an admin to go first.
- Org admins control which clients may connect. By default, Atlassian-supported domains are allowed automatically. Admins can block them — but it’s all-or-nothing, individual domains can’t be blocked selectively — and can authorize custom domains.
- API token auth is off until an admin turns it on. Tokens are the headless path and are required for Jira Service Management Ops and Bitbucket Cloud, since OAuth doesn’t cover those. An org admin enables it under Atlassian Administration → Rovo → Rovo MCP server → Authentication.
Tool invocations are written to audit logs and the server honors Cloud IP allowlists, which is usually what security review wants to hear.
On plan eligibility: Atlassian’s MCP documentation doesn’t state it anywhere. Third-party posts claim Standard and above with no Free tier, and that Rovo needs a verified business domain — plausible, but not confirmed by an Atlassian page, so verify against your own site rather than a blog. Atlassian does state that the MCP server itself doesn’t consume Rovo credits.
The rate limits nobody will quote you
Atlassian publishes no rate-limit numbers for the MCP server. Not in the troubleshooting doc, not in the developer portal, not in the GA announcement, not in the repo. If you see a specific figure like “1,000 requests per hour” cited confidently, be skeptical — the most-linked source for that number attributes it to a README that doesn’t contain it.
What is real is the friction. On GitHub issue #171, a user reported 429s after roughly 20 parallel calls and only 200-300 calls across a couple of hours, with no maintainer response. A community thread describes a single query returning 30 cards, then being blocked “for at least 30 minutes.” Another user notes the retry-after header information isn’t passed through, so the model has no hint how long to wait and just keeps failing. Reports were still arriving four days ago.
Atlassian’s only published mitigation is a tip: set maxResults: 10 on JQL searches and limit: 10 on CQL. Take that seriously — an agent that runs a broad JQL query and then fans out per-issue calls is the exact pattern that gets you throttled.
Nothing fires on its own
There is no such thing as an MCP trigger. The protocol's own working group states plainly that clients find out about server-side changes by polling for them, and a real event mechanism remains unshipped.
Every tool in Atlassian’s catalog is an imperative verb your client invokes. There is no listener, no subscription, no “when this issue transitions” hook. That’s the protocol, not an Atlassian shortcoming — but the practical consequence is the one that matters when you’re planning work.
It means the Jira MCP server can’t be the foundation of anything automatic. A ticket moving to In Review, a P1 getting filed at 2am, a sprint ending with eight issues still open — none of those can start anything. You have to open a chat and ask. Which is fine for “summarize what my team shipped this sprint,” and useless for “tell me when the release-blocking bug moves.”
That’s the gap Carly fills, and it’s a different job rather than a competing one. Carly connects to Jira natively — no server to point at, no OAuth scopes to reason about — and its workflows are event-triggered and scheduled, so they run without anyone in a chat window. When a bug is filed with priority Highest, post it to the on-call Slack channel and create the incident doc. Every Friday, summarize unresolved tickets per assignee and email the leads. The connecting-and-routing steps are free and unlimited, Zapier-style; the AI steps (drafting, summarizing, deciding) start at $35/month. Use the MCP server to interrogate Jira; use a workflow when Jira should be the thing that speaks first.
Community servers, and when they beat the official one
sooperset/mcp-atlassian is the dominant third-party option — 5.6k stars, actively maintained, with a release shipped this week and around 98 tools versus roughly 62 across all products on Atlassian’s server. People choose it for four reasons:
- Data Center and Server support. Confluence Server/DC 6.0+ and Jira Server/DC 8.14+. For self-hosted shops this isn’t a preference, it’s the only option.
- PAT and API-token auth. Skips the OAuth consent and admin-approval sequence entirely.
- Self-hosting. Runs Dockerized inside your own network, which is what air-gapped and compliance-constrained environments need.
- No Atlassian-side throttling. You hit your own instance’s REST API, which sidesteps the 429 problem above.
The tradeoff is ownership: hosting, patching, and the security posture are yours. For Jira Cloud with normal admin access, the official server is the better default — it’s maintained by Atlassian, it’s audit-logged, and it respects your permission model without you configuring anything.
If you’re wiring Jira into a specific assistant rather than evaluating MCP generally, Codex MCP servers and ChatGPT MCP servers cover client-side setup, and connecting Jira to an AI agent covers the non-MCP paths.
FAQ
Does Jira have an official MCP server?
Yes. The Atlassian Rovo MCP Server went generally available on February 4, 2026, and is hosted by Atlassian at https://mcp.atlassian.com/v1/mcp/authv2. It covers Jira, Confluence, Jira Service Management Ops, Bitbucket Cloud, and Compass.
Does the Atlassian MCP server work with Jira Data Center or Jira Server?
No. Atlassian states the official server is “designed and supported only for Atlassian Cloud products” and that an active Atlassian Cloud site is a prerequisite. There is no published timeline for Data Center support. Self-hosted instances need a community server such as sooperset/mcp-atlassian, which supports Jira Server/DC 8.14+.
What can the Jira MCP server do? Sixteen Jira tools: JQL search, reading issues and their metadata, creating issues, editing fields, transitioning status, adding comments, and logging work. There is no delete-issue tool.
Can the Jira MCP server react to an issue changing status? No. MCP is request/response — there are no triggers, no scheduler, and no webhooks. An AI can transition an issue when you ask it to in a chat, but nothing fires automatically when a ticket moves, a bug is filed, or a sprint ends. Event-driven Jira work needs a workflow tool rather than an MCP server.
Why am I getting rate-limited by the Atlassian MCP server?
Atlassian doesn’t publish rate-limit numbers, but users report 429s after a few dozen calls, sometimes with 30-minute lockouts, and retry-after information isn’t passed back to the model. Atlassian’s guidance is to cap searches with maxResults: 10 for JQL and limit: 10 for CQL.
My Jira MCP connection stopped working — what changed?
Most likely your client is pointed at https://mcp.atlassian.com/v1/sse. That endpoint was deprecated in March 2026 and switched off on June 30, 2026. Move to https://mcp.atlassian.com/v1/mcp/authv2.
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
