BusinessRnD · draft v2 · built 2026-10-21 06:15 UTC
I12 · Atlassian Forge apps — deep dive
Side bet
Best fees anywhere, crowded shelf, payout to Ukraine unconfirmed.Realistic: $1–4k/mo after 12 months · Chance of $10k/mo: low
Cards have Go deeper, Maybe and Kill buttons. Press Send feedback to save. Your marks from all tabs collect at the end of each page.
Bottom line
The money terms are the best of any channel, but Ukraine payout is not confirmed yet.
Atlassian takes 0% of Forge revenue until you have earned $1M lifetime (pricing).
It bills customers, collects tax and pays USD to "a valid bank account in your name" once $500 is owed (agreement).
No public page lists payout countries, so pass vendor verification before building app #2.
The shelf is crowded, not empty.
There are 6,696 Jira and Confluence cloud apps. The median one has 6 installs (Jira) or 16 (Confluence).
About 2,230 apps listed since mid-2025 have a median of 1–2 installs, and only one reached the Jira top 200 by revenue (Marketplace API, 20 Oct 2026).
Money is concentrated.
About 1,400 Jira and 500 Confluence apps earn anything (estimate).
Comparable apps charge $0.10–2.92 per user per month, with a median of $0.90.
22 of 35 comparable apps are free up to 10 users.
The openings have dates on them:
293 paid Connect apps that still earn have been frozen since March 2026 and lose support on 31 Jan 2027.
Data Center sales to new customers ended on 30 Mar 2026, and DC goes read-only on 28 Mar 2029.
The two most-installed free Mermaid renderers (11,756 installs) are rated 2.8–3.0★.
People buy, agents use.
Apps sold as "a Rovo agent" have 20–90 installs.
But apps whose actions customer agents can call stand out: only 270 of about 6,700 cloud apps are in the Rovo collection.
The customer pays for that reasoning.
Build these three first:
Diagram Blocks: renders the Mermaid and Markdown that agents write.
Leak Guard: a secrets scanner where nothing leaves Atlassian.
API Shelf: an OpenAPI viewer replacing frozen Connect apps.
None of them calls an AI model at runtime. Each keeps the "Runs on Atlassian" badge and takes the AI 1–3 weeks to build.
Money (estimate):
$1–4k a month from the top three after 12 months, or $8–15k if they hit the upside.
$10k a month needs 4–6 apps and 12–30 months.
Kill any app with fewer than 15 trials by day 30 or 3 paying sites by day 60.
What the Marketplace data says
On 20 Oct 2026 I pulled from Atlassian's public REST API:
every cloud app: 4,820 for Jira and 1,876 for Confluence;
two rankings: by revenue ("top-grossing") and by momentum;
all Data Center listings;
19,773 recent reviews;
live prices for 35 comparable apps.
"#44" means 44th by revenue in that product. Installs include Data Center installs for apps sold on both. The files and the method are in the research notes.
Demand and supply by category
Category
Apps
Installs
Strong paid apps*
In top 200 by revenue
Jira · Project management
1,161
610k
94
122
Jira · Workflow
598
427k
42
49
Jira · Reports
563
367k
60
70
Jira · IT & helpdesk (JSM)
521
227k
33
51
Jira · Admin tools
480
195k
21
40
Confluence · Macros
669
357k
46
112
Confluence · Documentation
492
243k
38
112
Confluence · Diagramming
222
228k
19
39
Confluence · Document management
325
124k
27
83
*Paid, 1,000+ installs, 4.0★ or better.
Power law. Only 230 Jira and 103 Confluence apps have 1,000+ installs.
Portfolio vendors dominate. Only 13 of the Jira top 100 come from vendors with three apps or fewer.
Revenue per install is highest in compliance scanning, document approval, config backup and "zero egress" editions. For example, draw.io Zero Egress ranks #104 with 209 installs.
Apps listed since mid-2025
Jira
Confluence
New apps
1,720
513
Median installs
1
2
500+ installs
15
1
In top 500 by revenue
13
6
In top 200 by revenue
1
0
Niches: demand, crowding, quality
Niche
Apps
Installs
Listed since mid-2025
Apps earning
Quality signal
Diagrams as code
153
81k
45%
49
Top free renderers rated 2.8–3.0★
Confluence approvals
151
25k
64%
39
Leader at 1.88★ recently
Time in status
113
60k
60%
36
Strong 4.5★ incumbents
OpenAPI viewers
33
21k
21%
17
3 Connect apps frozen
Jira admin cleanup
96
12k
60%
26
DC-only tools, 7k installs
Release notes
28
10k
43%
10
No new app above 100 installs
JSM surveys
47
9k
23%
17
Top five average 3.64★
Secrets and personal data
65
6.5k
37%
16
High price per install
Glossary
17
2.5k
41%
4
Overlaps Rovo Definitions
Read confirmation
15
0.6k
87%
3
9 clones in four months
Chart positions come from the data:
Demand blends installs and the number of earning apps.
Competition blends the share of new apps, strong incumbents and app count.
Agent analytics uses adjacent demand (installs of coding-agent apps), because it has no incumbents.
quadrantChart
title Niche demand vs competition
x-axis Low competition --> High competition
y-axis Low demand --> High demand
quadrant-1 Big but crowded
quadrant-2 Best shelf
quadrant-3 Small and open
quadrant-4 Avoid
Diagrams as code: [0.69, 0.97]
OpenAPI docs: [0.27, 0.74]
Secrets and PII: [0.31, 0.66]
Jira config admin: [0.55, 0.78]
Recycle bin: [0.13, 0.52]
Page approvals: [0.77, 0.92]
Release notes: [0.32, 0.65]
Agent analytics: [0.05, 0.35]
Glossary: [0.20, 0.51]
JSM CSAT: [0.31, 0.69]
Time in status: [0.80, 0.95]
Planning poker: [0.47, 0.75]
Read confirm: [0.38, 0.42]
xychart-beta
title "Share of niche apps listed since mid-2025 (%)"
x-axis [ReadConfirm, Approvals, JiraAdmin, TimeInStatus, Diagrams, ReleaseNotes, Secrets, Poker, CSAT, OpenAPI]
y-axis "Percent" 0 --> 100
bar [87, 64, 60, 60, 45, 43, 37, 37, 23, 21]
What users complain about
Keywords in 3,068 one- and two-star cloud reviews: bugs came up 553 times, missing features 397, support 315 and price only 72. Quality and support win customers; discounts don't. Examples:
Mermaid Chart: tiny images, unreadable large diagrams.
Render Markdown: frozen on Connect, and its output isn't searchable.
Page Approval: 1.88★ since it switched from free to paid.
Two dated gaps: Connect freeze and Data Center end of life
Connect apps can't be updated since March 2026, and support ends on 31 Jan 2027 (Atlassian). In the data, 208 Jira and 85 Confluence paid Connect apps still earn. Apps with no update for 18 months may be hidden (changelog).
Data Center: new customers can't buy DC after 30 Mar 2026, and DC turns read-only on 28 Mar 2029 (Atlassian FAQ). Most DC vendors already relisted for cloud. The clearest uncovered demand is admin tooling:
Ranking uses keyword match, a meaning match on the app's name and tagline, and engagement: reviews, rating, installs and support details (ranking). Trials last 30 days, the app's tier matches the site's user count, and annual price is 10 times monthly.
Agents reach an app only through its actions:
the app's own Rovo agent;
agents customers build in Rovo Studio, via the rovo:mcp module;
external clients like Claude, Cursor and Codex, once a site admin switches that on. It is off by default and still in preview (rovo:mcp).
"Marketplace partners are not responsible for the AI usage costs of their Rovo agents" (Forge pricing). Since July 2026, teams can assign Jira work to Claude, Cursor and Copilot (Atlassian), so agents now write much of Jira and Confluence.
flowchart TD
P1["Admin or user searches Marketplace or Find new apps"] --> P2["Ranking by keywords, meaning, reviews, installs"]
P2 --> P3["30-day trial or user request to admin"]
P3 --> P4["Paid subscription billed by Atlassian"]
A1["Customer Rovo agent built in Studio"] --> A3["Your app actions exposed through rovo mcp"]
A2["Customer Claude Code, Cursor or Codex"] --> A5["Site admin enables external MCP"]
A5 --> A3
A4["Your app's own Rovo agent"] --> A3
A3 --> A6["Customer Rovo credits pay for reasoning"]
A3 --> A7["You pay only Forge compute, mostly free tier"]
A3 --> P5["Daily agent use keeps the subscription renewing"]
P5 --> P4
So:
Sell to people.
Expose 3–8 narrow actions so customer agents use the app daily.
Skip agent-only products: nobody buys them yet.
The economics and rules
Item
What applies
Source
Revenue share
100% of gross until $1M lifetime Forge revenue (since 1 Jan 2026), then 83%. The app needs no Connect modules, OAuth and Forge UI
Free per app per month: 200,000 GB-seconds plus small storage. Overage is tiny. Pay by card or PayPal. An unpaid bill disables all your apps after 14–21 days
With a 100% share and about 6% FOP tax, $10k a month in profit needs about $10.7k in gross sales. Cash comes after the trial, the $500 threshold and payment about 30 days after month end. Expect the first payout 2–4 months after launch.
Concrete plugins
I ranked by:
revenue evidence inside the Marketplace;
gap size;
crowding;
build effort;
no-model fit;
agent use;
risk that Atlassian builds it natively.
#
Plugin
Product
People or agents
Strongest evidence
Build
1
Diagram Blocks
Confluence + Jira
Both
81k niche installs; leaders below 3★
1–2 wk
2
Leak Guard
Confluence + Jira
Both
#51 with 248 installs
2–3 wk
3
API Shelf
Confluence
Both
4.7k installs on frozen apps
1–2 wk
4
Config Inspector
Jira
People first
DC-only tools, 7k installs
3–5 wk
5
Recycle Bin for Jira
Jira
Both
Frozen incumbent at #287
1 wk
6
Simple Approvals
Confluence
People
39 earning; leader at 1.88★
2–3 wk
7
Release Notes Desk
Jira + Confluence
Both
Incumbent at #108
1–2 wk
8
Agent Desk
Jira
People
No incumbent; 17k agent installs
1–2 wk
9
Term Base
Confluence
Agents first
#36 and #68
1–2 wk
10
Pulse CSAT
JSM
People
Top five at 3.64★
2–3 wk
Open the mockup — Diagram Blocks in Confluence and Leak Guard in Jira, with fictional data.
Diagram Blocks — Mermaid and Markdown for Confluence and Jira
What it is: A macro and panel that render Mermaid and Markdown as sharp, full-width diagrams (SVG), with zoom, full screen, dark mode and export. It turns pasted Mermaid code blocks into diagrams, either with one click or automatically.
Who uses it (people or agents): Both. People read and edit. Agents (Claude Code writing through Atlassian's MCP server, or Rovo) write the Mermaid and call insert_diagram.
Evidence of demand: 153 apps, 81k installs, 49 earning. Stratus Mermaid ranks #44 at $0.40 per user, PlantUML #10 and Just Add+ #16.
The gap: The top two free renderers (11,756 installs) are rated 2.99★ and 2.84★. The Markdown leader is frozen on Connect.
Why we can win: Fix the named complaints, ship agent actions on day one, keep zero egress and win on reviews. Risks: 69 apps entered since mid-2025, and Atlassian could build a native renderer.
Build effort (AI): 1–2 weeks.
Compute: No model. Diagrams render in the reader's browser.
Pricing & revenue estimate: Free up to 10 users, then $0.35 per user per month.
Base: $500–1,500 a month at 12 months.
Upside: $4–6k.
First step: Build v1 against the most-cited complaints.
I12.2
Leak Guard — secrets and personal-data scanner with zero egress
What it is: Scans new and edited work items, comments and pages for keys, passwords, tokens, card numbers and personal data. It alerts the author, masks text on request, keeps an audit log and can scan old content once. Nothing leaves Atlassian.
Who uses it (people or agents): Both. Admins review findings. Agents call check_text_for_secrets before posting, and the scanner catches logs that agents paste.
Evidence of demand:
Soteri: Confluence #51 with 248 installs, at $1.74 per user.
24 apps listed since mid-2025 have 192 installs between them.
Agents create a new way for secrets to leak.
Why we can win: Runs on Atlassian as the pitch, at about one-eighth of Guard Premium ($8.55 per user, includes scanning). Risks: buyers want bug-bounty trust signals, and Guard is native.
Build effort (AI): 2–3 weeks, then ongoing rule tuning.
Compute: No model: pattern rules and checksums. Under $5 a month of Forge compute (estimate).
Pricing & revenue estimate: $1.00 per user per month.
Base: $500–1,500 a month.
Upside: $4–7k.
First step: Confluence version first, with a public list of what it detects.
I12.3
API Shelf — OpenAPI docs in Confluence that coding agents can read
What it is: Renders OpenAPI and AsyncAPI specs stored in Confluence and searches across them. A "replace old macro" action swaps out frozen Connect macros. There are no live "try it" calls, because those would send data out.
Who uses it (people or agents): Both. Developers read the docs. Coding agents and Rovo call search_endpoints and get_operation.
Evidence of demand: 33 apps, 21k installs, 17 earning. Three frozen Connect viewers hold 4,726 installs, led by Open API (Swagger) Integration at #115.
Only 7 apps entered since mid-2025, the least crowded niche I checked.
Why we can win: A dated switching moment, and internal-API lookup for coding agents, which nobody offers.
Build effort (AI): 1–2 weeks.
Compute: No model. Docs render in the browser.
Pricing & revenue estimate: $0.25 per user per month. Incumbents charge $0.10–0.15.
Base: $300–800 a month.
Upside: $2–3k.
First step: List with a switch guide for users of the frozen apps.
I12.4
Config Inspector — Jira Cloud cleanup for migrated sites
What it is: Shows where each custom field, status, screen and scheme is used, what is unused, and a safe cleanup plan.
Who uses it (people or agents): People first: admins and migration partners. Agents ask where_is_field_used.
Evidence of demand: DC-only Integrity Check (2,810 installs) and Power Admin (2,598). In cloud, Optimizer ranks #126 and Doctor Pro #215 at $1.80.
The gap: The DC end of life moves cluttered setups to cloud. 58 new admin apps have 159 installs combined, so there is no winner yet.
Why we can win: Priced on all site users, and the checks are rule-based. Risks: Jira already shows "last used" for fields, and large sites hit API limits.
Build effort (AI): 3–5 weeks.
Compute: No model. Scheduled scans cost a few dollars a month at most (estimate).
Pricing & revenue estimate: $0.60 per user per month.
Base: $500–1,500 a month.
Upside: $3–7k.
First step: A read-only field-usage view.
I12.5
Recycle Bin for Jira — undo for deleted work items
What it is: Snapshots deleted work items for 30–90 days. Admins can restore one as a new linked item.
Who uses it (people or agents): Both. It gives agents that delete through the MCP server an undo, via restore_item.
Evidence of demand:Restore Deleted Issues ranks #287 at $0.30 per user. It is frozen on Connect.
The gap: The incumbent is frozen, only 3 apps earn, and agent write access raises the risk of accidental deletes.
Why we can win: Tiny build, little competition.
Build effort (AI): About 1 week.
Compute: No model. Snapshot storage costs cents per customer.
Pricing & revenue estimate: $0.30 per user per month.
Base: $200–600 a month.
Upside: $1.5–3k.
First step: Confirm Jira has no native restore.
I12.6
Simple Approvals — page approval and bulk read-confirm for Confluence
What it is: Draft, review and approved statuses, named approvers, locking after approval, and read-confirm campaigns across many pages at once.
Who uses it (people or agents): People. Agents use my_pending_approvals.
The gap: 12 new apps since mid-2025, none above 100 installs.
Why we can win: Zero egress, and the AI writing costs you nothing. Risk: Rovo alone can draft notes.
Build effort (AI): 1–2 weeks.
Compute: No model on your side. The customer's Rovo uses about 10 credits per run.
Pricing & revenue estimate: $0.40 per user per month.
Base: $200–700 a month.
Upside: $2k.
First step: Reuse the Diagram Blocks template engine.
I12.8
Agent Desk — scorecard for work assigned to coding agents
What it is: A Jira gadget tracking work assigned to Claude, Cursor or Copilot: throughput, cycle time compared with humans, reopen rate and how long work waits for human review.
Who uses it (people or agents): People. Agents ask agent_scorecard.
Evidence of demand: No incumbent. The adjacent apps are big: Claude Agent for Jira has 9,629 installs, Cursor 3,727 and Copilot 3,544, all within about three months of launch (collection).
The gap: Teams can't see in Jira whether agent work actually ships.
Why we can win: First mover. Risks: willingness to pay is unproven (round 3), and Atlassian announced DX AI cost management.
Build effort (AI): 1–2 weeks.
Compute: No model.
Pricing & revenue estimate: $0.50 per user per month.
Base: $0–500 a month.
Upside: $3k.
First step: List it free; charge only at 100+ installs.
I12.9
Term Base — governed glossary for people and agents
What it is: An approved glossary with owners and review dates, a "terms on this page" panel, and actions for agents.
Who uses it (people or agents): Agents first, through lookup_term.
gantt
title First 20 weeks
dateFormat YYYY-MM-DD
section You once
Atlassian account and partner enrolment :o1, 2026-10-21, 3d
Domain email, verification, payout, card :o2, 2026-10-22, 10d
section Build and list
Diagram Blocks v1 :b1, 2026-10-21, 12d
Diagram Blocks review :r1, after b1, 15d
Leak Guard v1 :b2, after b1, 15d
Leak Guard review :r2, after b2, 15d
API Shelf v1 :b3, after b2, 10d
API Shelf review :r3, after b3, 15d
section Measure
Day 30 and day 60 checks :m1, after r1, 60d
Portfolio decision :milestone, m2, 2027-03-15, 0d
I do:
Build, test and deploy each app with Atlassian's Forge MCP server.
Write the listings, docs, EULA and privacy drafts, and the Privacy & Security tab.
Handle the review rounds.
Track installs, trials, reviews and frozen competitors weekly.
Draft support replies and security fixes within Atlassian's deadlines.
You do once:
Create the Atlassian account, Developer Space and partner enrolment, with two-step verification.
Pass verification: FOP details, passport and selfie.
Set up a company-domain email.
Add USD payout details and the tax form.
Add a card or PayPal for Forge billing.
Approve the EULA and privacy policy.
You do recurring:
Near zero: rare escalations, like signing a customer security questionnaire.
Your usual quarterly FOP declaration.
Go and kill (per app):
Day 30: at least 15 trials or installs. Otherwise fix the listing once, then stop.
Day 60: at least 3 paying sites or $100 a month. Otherwise freeze the app.
Day 120 (portfolio): $1,000+ a month across the apps, or stop building new ones.
Any time: if Ukrainian payout fails, pause everything. If Atlassian ships the feature natively, freeze that app.
Risks and open questions
Ukraine payout is unverified. There is no public country list, the forms sit behind a login, and Stripe Identity is unconfirmed for Ukrainian passports. Expect currency-control questions from your bank.
Clones: 21–87% of each niche's apps arrived since mid-2025. Reviews and speed are the moat.
Atlassian builds it natively: Guard already scans, Rovo Definitions overlap glossaries, and native Mermaid could follow.
Policy: the 100% share can change on 30 days' notice, and an unpaid Forge bill disables every app you have.
Trust ceiling: security buyers may want a bug bounty and the AEC badge.
Unknowns: revenue behind each rank, trial-to-paid conversion and average site size.
Open questions:
Does Jira restore deleted work items natively?
When does external MCP access become generally available?
Do apps with only rovo:mcp actions join the Rovo collection?