mcp-jira — VS Code Copilot ⇄ Jira
MCP Server for In-IDE Ticket Workflows
Context
MCP-Jira is the ticketing-context tool layer that lets a Claude agent read and write tickets without leaving the IDE — same MCP-as-integration-layer pattern as MCP-SQL, applied to a different system of record. The architectural call: every agent action against Jira flows through one auditable server with guardrails (backlog placement enforced, missing transitions return the available list instead of failing silently, file existence validated before attachment) instead of through nine ad-hoc HTTP calls scattered across whatever script needs Jira context. Atlassian Document Format (ADF) builder transparently converts the agent's plain text into Jira's required rich-text JSON. Credentials resolved from OS keychain or env vars; never hardcoded. In the Modernization Factory, MCP-Jira is the tool layer agents call when a translation step needs ticket context (acceptance criteria, prior decisions, linked epics) to make a defensible call. Saves an estimated 15–20 minutes per developer per day on routine ticket work.
Architecture
What I owned
- The architectural call: one auditable MCP boundary for agent → Jira instead of ad-hoc HTTP calls scattered across scripts
- Component of the Modernization Factory: the tool layer for ticket-context retrieval when an agent needs acceptance criteria or linked-decision history
- 9 tools covering create / read / update / search / transition / comment / attach across Jira's core workflow
- ADF builder auto-converts plain text to Jira's rich-text JSON (heading/list/paragraph heuristics)
- Guardrails: backlog placement enforced; missing transitions return the available list rather than failing silently
- Credentials via OS keychain (Windows Credential Manager / macOS Keychain) or env vars — never hardcoded
- Stdio JSON-RPC transport — launched as a subprocess by VS Code, zero additional infrastructure