Jira attachments

The official Atlassian MCP returns attachment metadata (id, filename, mimeType, a content URL) but has no tool that returns the bytes — and its OAuth token can't be borrowed. This skill closes that gap: metadata comes from the MCP (existing auth), and the binary is downloaded with a personal Atlassian API token, then read from disk so the image is genuinely viewable (not just base64 text).

Prerequisites (one-time)

The download step needs these in the shell profile (~/.zshrc or a sourced file):

export ATLASSIAN_EMAIL="you@example.com"          # email on your Atlassian account
export ATLASSIAN_API_TOKEN="…"                     # id.atlassian.com/manage-profile/security/api-tokens
export ATLASSIAN_SITE="your-site.atlassian.net"    # your Jira Cloud site host

If a download fails with an auth error, the token/email/site are missing or wrong — stop and tell the user to set them; do not go hunting for credentials elsewhere.

Steps

1. Resolve the ticket. Accept a ticket key (e.g. ABC-1234) or an issue URL. Determine the cloudId for the site by calling getAccessibleAtlassianResources (do not hardcode it — it differs per site).

2. Get attachment metadata via the MCP (no token needed): getJiraIssue(cloudId, issueIdOrKey, fields:["summary","attachment"]). Each attachment gives id, filename, mimeType, size, content.

   • If attachment is empty, tell the user the ticket has no attachments and stop.
   • If the user named a specific file, filter by filename; otherwise take all image/* attachments (offer to include non-images too).

3. Download each wanted attachment into the session scratchpad using the helper — pass the attachment id (the script builds the correct site-URL form itself):

   bash ~/.claude/skills/jira-attachments/fetch-attachment.sh <attachment_id> "<scratchpad>/<filename>"
   

   Use the real session scratchpad directory for <scratchpad>. The script prints saved: <path> (<mime>, <bytes>) on success and exits non-zero with the response body on failure.

4. View it. Read each saved file — image files render visually — then answer the user's actual question (describe the UI, extract the error text, compare to code, etc.).

Notes

• Download uses https://<site>/rest/api/3/attachment/content/<id> with Basic auth. Do not use the content URL from metadata verbatim — it points at api.atlassian.com/ex/jira/..., which only accepts OAuth Bearer, not the API token.
• Only the byte download uses the token; all metadata stays on the MCP.
• Files land in the scratchpad, so no cleanup is required and nothing enters the repo.