Workato
Azure
MuleSoft
Salesforce
AWSReady To Modernize Your Architecture?
Discuss your current infrastructure and scalability goals with our integration experts.
Article Series
From Clean Slate to Working PatternPart 2 · you are here
In my last post, I talked about why I built a read-only Confluence MCP server on Workato instead of using Atlassian’s own connector: I wanted to own the control plane, not just accept a vendor’s default.
This post is about something that surprised me while building it: I’m a practice director, not a developer, and I built the whole thing myself. No engineer in the loop, no ticket, no waiting.
If you're wondering who's actually capable of building governed AI tooling inside your organization, the answer might be a wider group of people than you think.
Starting from a question, not a spec
I didn’t sit down with a build plan. I sat down with a real business question: how current and complete were our own documented Workato standards, actually, ahead of a client delivery?
That’s not a developer question. It’s a practice-management question. But answering it well meant reviewing a large, messy Confluence space by hand, which doesn’t scale.
So I described what I needed to AIRO, Workato’s AI build agent, in plain language. Three read-only skills to start:
Search
find pages by keyword across the space
Retrieve
pull full page content by ID
List
enumerate pages in a given space or folder
I didn’t write connector code. I described intent, and AIRO generated the underlying actions and recipe logic. That’s the part worth sitting with: the barrier between “I have a question about our documentation” and “I have a working tool that answers it” got a lot shorter, and it didn’t require pulling a developer off other work to close that gap.
Where the question kept pushing back
The three starting skills weren’t the whole answer. Confluence’s API will hand you a folder’s numeric ID without blinking, but it won’t hand you the folder’s actual title in the same call. If you’re clicking through Confluence in a browser, you never notice, the UI resolves it for you. If you’re trying to build a real folder hierarchy programmatically, you hit a wall: a tree of IDs and no way to know what any of them are called.
I noticed the output wasn’t right, a folder name coming back as a numeric ID instead, and asked AIRO about it. Its response pointed me toward building a dedicated resolution step rather than patching the symptom inline. That became a fourth skill, Get Folder Names, a standing part of the server rather than a one-off fix.
No ticket to another engineer. No waiting on someone else’s availability. I hit a wall, described the problem in plain language, and worked through it with the tool right there. That’s the pattern I want a reader to take away: the skillset that used to gate this kind of build, hand-rolling API logic, debugging integration quirks, isn’t the gate anymore. Knowing your data and knowing what question you’re actually trying to answer is.
Worth noting
AIRO is still in beta as I write this, with general availability looking close, though I don’t have a firm date. It’s worth checking on Workato’s end before you plan around it, but if this is the state of things in beta, that’s itself worth paying attention to.
Four Skills, Built by Describing Them
Three to start. A fourth because the question pushed back.
Skill 01
Search
Find pages by keyword across the space.
Skill 02
Retrieve
Pull full page content by ID.
Skill 03
List
Enumerate pages in a given space or folder.
Skill 04 · The Gap
Get Folder Names
Resolve numeric IDs into real folder titles — a standing part of the server, not a one-off fix.
184
page Confluence inventory, with a readable folder hierarchy
0
engineers in the loop, tickets, or lines of connector code
2
workstreams scoped in the consultant work-assignment workbook
The payoff
Four skills, one resolved gap in the folder-to-name mapping, and the result was a 184-page Confluence inventory with a real, readable folder hierarchy instead of a flat list of IDs. From there it was a short step to a refactor action list flagging what was stale or missing, and a consultant work-assignment workbook splitting the remaining review into two workstreams.
That’s a real business deliverable, produced by describing a problem clearly and staying with it until it was solved, not by writing custom software.
The part I’m not going to oversell
I built this using a personal, scoped, read-only token, which is the right instinct: no shared credential, no write access, every call traceable to me. I’ll be straight about the weak point too. That token currently lives in the chat context I used to build and run this, which is a weaker storage model than a proper secrets vault. The read-only scope limits how much damage exposure could do, but it doesn’t eliminate the exposure risk itself.
Workato has Verified User Access coming for this kind of setup, which should close this gap by tying each call to a live, runtime authentication flow instead of a pre-issued token sitting in a chat. I’d rather tell you where the current approach is still maturing than pretend it’s finished. That honesty is part of the point: this is a genuinely new way of building, and it’s still catching up to itself in places.
Who controls the plane. Who can build on it. Who governs it in production.
The bigger question this raises is what happens when a tool like this runs against real documentation with real sensitive data in it, not just a standards review. My next post covers exactly that: what Agent Guardrails actually caught mid-review without me writing a rule for it, and how governance holds up once you’re running this in production.
If Post 1 was about who controls the plane and this post was about who can build on it, Post 3 is about what happens when you actually govern it in production.
If you’re trying to figure out how far a non-engineer on your team could get with AI-assisted building, or what it would take to run something like this safely inside your own systems, that’s the conversation API People is built to have. Reach out to us to talk through what this looks like for your environment.
Reach out →FAQs
What is AIRO?
AIRO is Workato’s AI build agent. Instead of writing connector code by hand, you describe what you need in plain language, and AIRO generates the underlying actions and recipe logic. It was still in beta at the time this server was built, with no firm general availability date announced.
Can someone without engineering experience really build an MCP server?
Yes — this Confluence MCP server was built by a practice director, not a developer, with no engineer in the loop, no ticket, and no waiting. The barrier between having a question about your documentation and having a working tool that answers it has gotten a lot shorter with AI build agents like AIRO.
What skills does this Confluence MCP server have?
Four read-only skills: Search (find pages by keyword across the space), Retrieve (pull full page content by ID), List (enumerate pages in a given space or folder), and Get Folder Names (resolve numeric IDs into real folder titles, added after the first three skills hit a gap in Confluence’s API).
Is it safe to use a personal access token to build tools like this?
The safer instinct is a personal, scoped, read-only token — no shared credential, no write access, every call traceable to one person. The current weak point is that the token lives in the chat context used to build and run the tool, which is a weaker storage model than a proper secrets vault, though the read-only scope limits how much damage any exposure could do.
What is Workato Verified User Access?
It’s an upcoming Workato capability that ties each call to a live, runtime authentication flow instead of relying on a pre-issued token sitting in a chat context — closing the credential-storage gap described above.
What did this project actually produce?
A 184-page Confluence inventory with a real, readable folder hierarchy instead of a flat list of IDs, a refactor action list flagging what was stale or missing, and a consultant work-assignment workbook splitting the remaining review into two workstreams — all with zero engineers in the loop.
Free PDF
Read it offline — get the full PDF
Fill in two fields and the download starts right here.
- Every figure and diagram at full resolution
- All sources, linked and cited
No spam. We respect your inbox and your data.
Contact Us
Talk to an integration expert
About your environment — no pitch deck required.
