Can Rovo be used as an M2M (server‑to‑server) backend, or is it limited to interactive use only?

[Mariano Tallarico _Appricity_]

Hi everyone,

We’re evaluating whether we can use Rovo as a “knowledge backend” behind an external chat (N8N + LLM), where the UI is ours but the Jira/Confluence context comes from Rovo.

From the docs, it looks like Rovo/MCP access is designed around interactive OAuth 2.x flows (browser login, user consent, short‑lived tokens) for clients like ChatGPT/Claude/IDEs, and we haven’t found any reference to:

• Rovo‑specific service accounts / service tokens / API tokens.
• OAuth ‎`client_credentials` or similar machine‑to‑machine flows.
• Official patterns for server‑to‑server integrations (e.g. from N8N or another backend).

Concrete questions:

1. Is there currently any supported way to use Rovo as a technical M2M service (via service accounts, API tokens, or ‎`client_credentials`), or is it intentionally limited to usage within the Atlassian ecosystem (UI, Forge, Automation, official Slack/extension, etc.)?
2. If not supported today, is there anything on the roadmap or any public feature requests to expose Rovo as an API consumable from external backends?

We’d really appreciate a clear confirmation so we can decide whether to keep trying with Rovo or build our own RAG/vector store instead.

Thanks in advance.

[Hernan Montes]

Hi Mariano,

I have been working with the MCP server connection myself and see a lot of things that still need to be fixed, and also, improve documentation.

To save you time, you need to follow Oauth flows, there's no direct token / user authentication against it.

Secondly, you will need to train your main LLM / AI to use the MCP, what the MCP server provides are just actions or tools to execute upon, so you will need to construct this library, ie. train your main model to use the MCP tools as needed.

Unfortunately, I don't see this MCP server as ROVO itself, it's just a set of integrations that will let you run actions against Jira - you can do the same with an API library.

[Ibby]

Hey @Hernan Montes 

Is it possible to share your list of improvements? your feedback is highly appreciated.

Thanks,
Ibby

[Hernan Montes]

Hi Ibraheem,

Glad to see tracking of these comments, I'm currently working an external solution, mainly for our internal use, and probably client use at some point, I worked with the Atlassian MCP server, constructed the auth flow and connected successfully, the only issue I see when doing it this way, is that we need to train our own AI to be able to use these tools, and I feel it still needs a lot of guardrails, like creating or updating tickets, if we give the tools absolute freedom to do so, we can end up in a destructive environment, we still use approve steps for every action the MCP should take.

As per your comment, I would like to see ROVO as an available external agent, this way, it can basically be guided by Atlassian own guardrails, rather than exposing it to users that may not be aware of the dangers of destructive changes.

I reconstructed the main needed MCP tools / actions into an API library, so basically a trained AI can make use of this library, and works like the MCP tools - but what people want is the ability to interact with the Atlassian ROVO at the teamgraph level, rather than exposing and getting Atlassian data out of the ecosystem, process it via external AI, then getting the data needed.

BIG PLUS - if you can let users create ROVO agents from within the Atlassian studio, and then let them reference them in outside integrations / code = game changer! IMO

Regards,

Hernan

[Ibby]

Thanks Hernan for sharing your thoughts and use cases :) that's exactly what we are thinking to do.

[Ibby]

Hi @Mariano Tallarico _Appricity_ 

Rovo MCP has Rovo Search tool today that can search content across Jira and confluence so this could be a good candidate to use it for RAG/Knowledge search thought the tool is free text based search not vector. MCP Server as you highlighted currently only supporting OAuth but we are exploring headless auth using API Tokens etc.

We are also exploring exposing Rovo as an Agent but I can share more details in few weeks about this.

Will Rovo Search tool in the MCP unblock you for now?

Thanks,
Ibby

[Georg Fankhauser]

HI @Ibraheem Osama , thanks for the information provided.

So do i get that right - the ROVO Search exposed through the Atlassian Rovo MCP Server is using a different technical baseline than ROVO Search within Atlassian products. You wrote that the former is using text based search not vector. Does that mean, that it is "just kind of an API wrapper" around standard API calls like issue search by JQL like also in the solution Hernan outlined - totally without the nice teams graph integration vector search etc.? I'd personally find the naming Rovo MCP Server a bit misleading then in the current status.

ROVO Search in the Atlassian Rovo MCP Server

Using Rovo search and fetch in the Atlassian Rovo MCP Server | Atlassian Rovo MCP Server Cloud | Atlassian Support

ROVO Search

Search | Rovo | Atlassian Support

Explore Atlassian Rovo: Search, Chat, and Agents | Atlassian

Being able to use Rovo Studio Agents from "outside" (supporting A2A?) would definitely be a good solutions (also independently) - I did not investigate, but is there a way to get that done now already?

[Ibby]

That's not what I meant. The Rovo Search tool in the MCP Server is using Rovo Search "That is exactly the same you use in product experience".

So in total the MCP Server has Rovo Search, JQL and CQL tools.

Sorry, what I meant by not support vector here, is the Rovo Search tool input is a string not an embedding. So Rovo search tool will be doing all the magic here similar to the Rovo Search in products.

[Ibby]

`but is there a way to get that done now already?`

Unfortunately not but it's on our radar :)