[Feature]: Enable reading Jira attachments directly via resource without download for remote MCP servers

[eric2788]

Is your feature request related to a problem? Please describe.

When running MCP as a remote server (HTTP/SSE), the agent cannot browse or access the download location of Jira attachments. This creates a significant limitation, as attachments need to be retrieved and processed, but direct file system access is unavailable to remote agents.

Describe the solution you'd like

Implement a feature that allows agents to read Jira attachments directly by passing a resource reference, rather than requiring the attachment to be downloaded to a local path. This would enable remote MCP servers and agents to access attachment data seamlessly, regardless of server location.

Describe alternatives you've considered

Currently, the only option is to download the attachment and provide the local path, which is not feasible for remote execution scenarios. No workaround exists that enables direct reading of the attachment resource.

Use Case

Remote MCP agents running over HTTP/SSE would be able to handle Jira attachments without the need for file system access, improving compatibility, usability, and automation for distributed environments.

Additional Context

This enhancement will help users running MCP in cloud or containerized setups, where filesystem access to downloaded attachments is not practical.

reference: github-mcp-server/get_file_contents

[jsguerrero]

Can you try workaround to mount volume on mcp server #449.

It's work for me

[eric2788]

Can you try workaround to mount volume on mcp server #449.

It's work for me

What if my docker server is remote? In that case, it seems that volume mounting might not function as expected.

[sooperset]

This is already implemented. jira_download_attachments returns attachment contents as base64-encoded EmbeddedResource objects over the MCP protocol — no filesystem access required on the server. This works for both local and remote MCP servers (HTTP/SSE/streamable-HTTP).

The tool never writes to disk; all data flows through MCP as embedded resources. This was the design from the start, specifically to support containerized and remote deployments.

[TanakornOmise]

I've use this jira_download_attachments to get image files. But AI can't able to see the Image and tell what the image is. Could you help me How can let AI read the file from base64-encoded.

[sooperset]

@TanakornOmise Thanks for reporting this! You're right — currently jira_download_attachments returns all attachments (including images) as EmbeddedResource with base64-encoded blobs. Most MCP clients treat these as opaque data, so the LLM can't actually "see" the image content.

On the Confluence side, we have confluence_get_page_images which returns images as ImageContent — the type MCP clients can render natively for LLM vision. We're missing the equivalent on the Jira side.

Tracked in #995 — we'll add a jira_get_issue_images tool that returns image attachments as ImageContent. No workaround available at the moment.