Jira REST API

Jira REST API Client

This skill documents the REST API fallback tier for the Jira integration. When the Atlassian MCP plugin is unavailable, commands use the jira_client.py script to interact with Jira directly via REST API v2 — except for issue search, which uses the v3 /search/jql endpoint after Atlassian removed the v2/v3 /search routes (HTTP 410, see CHANGE-2046).

Architecture

Phase 0 Cascade:
  1. Try MCP → JIRA_MODE = "MCP"
  2. Try REST → JIRA_MODE = "REST"
  3. Fallback → JIRA_MODE = "OFFLINE"

Agents are completely insulated — they produce identical output regardless of JIRA_MODE. Only commands and this script interact with the transport layer.

Script Location

Locate the script before invoking:

1. Glob: **/agent-team-creator/scripts/jira_client.py
2. Fallback: ~/.claude/plugins/agent-team-creator/scripts/jira_client.py
3. Not found → REST mode unavailable, fall to OFFLINE

Invocation Syntax

python3 {SCRIPT_PATH} \
  --action {action-name} \
  --config .claude/jira-rest-config.json \
  [--issue-key PROJ-123] \
  [--project PROJ] \
  [--query "search term"] \
  [--payload-file .claude/tmp/jira-payload.json] \
  [--file-path /abs/path/to/file] \
  [--no-cascade]

Credentials are ALWAYS read from the config file. Never pass tokens as CLI arguments.

Action Reference

Notes on search-issues

• Endpoint: POST /rest/api/3/search/jql (the v2/v3 /search routes were removed by Atlassian).
• total is an approximate count fetched from /rest/api/3/search/approximate-count in a separate non-fatal call. If that secondary call fails, total falls back to len(issues) so the search still returns useful data.
• Pagination is token-based, not offset-based. When more results exist beyond maxResults, the response includes nextPageToken; pass it back in the next payload's nextPageToken field to fetch the next page. There is no startAt.
• Default maxResults is 5; payload may override (Jira caps at 100).

Payload pattern: create-issue

{
  "project_key": "GCI",
  "issue_type": "Task",
  "summary": "Short title",
  "description": "Markdown body. Converted to wiki markup automatically.",
  "labels": ["claude-code", "automation"],
  "priority": "Medium",
  "assignee_account_id": "712020:abc-...",
  "parent_key": "GCI-40"
}

• project_key, issue_type, summary are required; everything else is optional.
• priority accepts the name as it appears in Jira (e.g. Highest, High, Medium, Low, Lowest, or custom names like P0–P3 if defined).
• assignee_account_id must be an Atlassian accountId (get yours via get-current-user).
• parent_key makes the new issue a child. Pair with issue_type: "Subtask" when the project's subtask type is named Subtask (call get-issue-types --project KEY to confirm the exact name).

Payload pattern: update-issue

{
  "summary": "New title (optional)",
  "description": "New markdown body (optional)",
  "labels": ["replaces", "existing", "labels"],
  "priority": "Low",
  "assignee_account_id": "712020:abc-...",
  "fields": { "duedate": "2026-12-31" }
}

• All top-level keys are optional but at least one must be set.
• labels replaces the existing label list (Jira's PUT semantics).
• Set assignee_account_id to null to unassign.
• fields is a raw escape hatch: any key/value here is merged into the Jira fields object verbatim. Use for fields the script doesn't model (e.g., duedate, customfield_10001).

Invocation pattern: attach-file

python3 {SCRIPT_PATH} \
  --action attach-file \
  --config .claude/jira-rest-config.json \
  --issue-key GCI-40 \
  --file-path /abs/path/to/report.md

• No payload file. The file is uploaded as multipart/form-data.
• Endpoint sends header X-Atlassian-Token: no-check (required by Atlassian for attachments).
• Markdown attachments display inline in the Jira UI; logs/screenshots/PDFs are linked.
• Multiple files require multiple invocations.

Invocation pattern: delete-issue

python3 {SCRIPT_PATH} \
  --action delete-issue \
  --config .claude/jira-rest-config.json \
  --issue-key GCI-40 \
  [--no-cascade]

• Default cascades subtask deletion (?deleteSubtasks=true).
• --no-cascade makes Jira refuse to delete an issue that has subtasks.
• This is irreversible — commands should confirm with the user first.

Notes on search-issues

• Endpoint: POST /rest/api/3/search/jql (the v2/v3 /search routes were removed by Atlassian).
• total is an approximate count fetched from /rest/api/3/search/approximate-count in a separate non-fatal call. If that secondary call fails, total falls back to len(issues) so the search still returns useful data.
• Pagination is token-based, not offset-based. When more results exist beyond maxResults, the response includes nextPageToken; pass it back in the next payload's nextPageToken field to fetch the next page. There is no startAt.
• Default maxResults is 5; payload may override (Jira caps at 100).

Exit Codes

Credential Configuration

Config file: .claude/jira-rest-config.json (MUST be in .gitignore)

{
  "baseUrl": "https://your-site.atlassian.net",
  "email": "your-email@company.com",
  "apiToken": "your-api-token",
  "configuredAt": "2026-02-24T10:00:00Z"
}

Generate API tokens at: https://id.atlassian.com/manage-profile/security/api-tokens

Security: API tokens have full account scope. The config file must NEVER be committed.

MCP-to-REST Parameter Mapping

Response Normalization

MCP and REST return different structures. Commands must normalize:

URL Construction

When JIRA_MODE = REST, construct issue URLs from the config:

{baseUrl}/browse/{issueKey}

Do NOT use MCP site resolution for URL construction in REST mode.

Temp File Pattern

For actions requiring --payload-file:

1. Create dir: .claude/tmp/ (if needed)
2. Write JSON: .claude/tmp/jira-payload.json
3. Invoke script
4. Parse stdout
5. Delete temp file (command responsibility)

Wiki Markup

REST API v2 uses Jira wiki markup (not markdown, not ADF) for description and comment body fields. The script converts markdown to wiki markup automatically. See references/wiki-markup-guide.md for the full translation table.

Rate Limiting

Jira Cloud: ~100 requests/minute per user. The script handles 429 responses automatically (parses Retry-After, sleeps, retries up to 2x). Commands typically make <10 requests per invocation.