Jira Rovo MCP plus Sume hosted MCP: turn a ticket into a release clip

You can connect Atlassian's Rovo MCP server and Sume's hosted MCP server to the same agent and let it turn a Jira work item into a short release clip: read the ticket through Atlassian, price the clip on Sume with dry_run, then submit it with a spend cap. The two servers never talk to each other. Your MCP client holds both connections, and the agent carries the text of the ticket from one to the other.

Atlassian's side comes from its getting started page for the remote MCP server, read on 2026-10-11. Sume's side comes from the MCP overview, tools and gates, OAuth and API keys and Jobs and results. Neither company documents a joint integration, so this is a pattern, not a product feature.

What does each server contribute?

Atlassian's page gives the endpoint as https://mcp.atlassian.com/v2/mcp and describes OAuth 2.1 as the primary sign-in, with API tokens as an optional alternative. It says the tools can summarize and search Jira, Confluence and Bitbucket, and create and update work items. Sume's endpoint is https://mcp.sume.com/mcp, and it adds generation, media inspection and job tools. Neither list contains the other's job.

What is the flow, step by step?

Start with the ticket. Ask the agent to read one work item through Atlassian and write a three-sentence release note from it. Keep the ticket text short; the agent should pass Sume a written prompt, not paste the whole ticket.

Then move to Sume. Have the agent call tools_schema for generate_video, then call it with dry_run=true. Sume's docs say dry_run returns an admission and cost preview and does not submit the job. If the estimate is acceptable, submit with a new idempotency_key and, if you want a hard ceiling, max_spend_usd. Omitting payload.model routes to sume/auto, per the docs.

Finally, poll with jobs_wait, which holds up to 55 seconds per call; on wait_slice_expired call it again with the same job id. Read jobs_result and hand the completed artifact URL back to Atlassian as a work-item update. Sume's core workflow page says completed jobs can carry public artifacts under https://media.sume.com, which is the URL to post, not a signed link.

Where do the credentials and approvals sit?

Atlassian's page tells users to apply least privilege, review high-impact changes before confirming, and watch audit logs. Treat the Jira write as one of those: let the agent draft the update and confirm it yourself. On Sume's side, the read-only OAuth scope is enough for the ticket-reading half of the day. Only the submit needs mcp:write or an API key, and Sume says there is no mcp:paid scope; the wallet and admission are the spend gate.

Keep the two credentials apart. A Sume OAuth token is not a Sume API key, and neither should be pasted into the ticket, a comment or the chat. If a key appears in a log or a Jira comment, rotate it.

What about gateways and many tools?

Atlassian's page notes that an MCP gateway that needs every tool should use the endpoint with ?tools=all. That is an Atlassian URL option, so it does not carry over to Sume. For Sume, tools_list is the source of truth for a session: with OAuth mcp:read it lists read-only tools, and with mcp:write or an API key it lists the paid and write tools too. A gateway that pins a tool list should be refreshed after you change the Sume scope.

One cost note keeps the flow honest. Atlassian charges Rovo credits per call, and Sume charges generation spend per job. Reading three tickets and generating one clip is two separate meters on two separate bills, and a retry on either side should reuse its own key rather than starting over.