MCP server for integrating with Redmine issue tracking system
RedmineMCP is a single-tool STDIO server with a minimal feature set. The one tool (list_issues) has a basic name and schema but lacks critical descriptions and output documentation. Tool naming follows the verb_noun convention correctly ('list_issues'), but parameter descriptions are insufficient, output schema is undocumented, and error handling provides no recovery guidance. The codebase shows raw JSON stringification of API responses without structuring, pagination support, or result limits. No input validation, no detailed error messages, and no security or audit patterns are evident.
Add a tool to list Redmine issues
Tool description is minimal (27 chars: 'Add a tool to list Redmine issues'). This description lacks context on when to call it, what structure is returned, and any usage constraints.
Parameter descriptions are generic and incomplete. 'limit' and 'offset' have descriptions ('Optional limit for number of issues returned' and 'Optional offset for pagination') but lack critical constraints: no stated min/max values, no indication of default behavior, no guidance on typical ranges (e.g., 'limit: 1-100, default 20').
Output schema is completely undocumented. The tool returns raw JSON from redmine.listIssues() wrapped in a generic text response, but there is no schema declaration of what fields are returned, their types, or structure. LLMs cannot plan downstream tool calls without knowing what data structure to expect.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 40 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 27 | - | v1 |
No pagination support or result limits enforced. The tool accepts limit/offset parameters but does not document a cap on maximum results. The underlying redmine.listIssues() call returns raw API response. Large unfiltered result sets blow context windows and degrade LLM reasoning.
Error handling is bare-bones. Errors return a generic 'Error listing issues: {message}' without actionable recovery guidance. Verify REDMINE_API_KEY is set' or 'Connection timeout. Retry in 30 seconds'). A raw error message gives the agent nothing to act on.
Credentials exposed via environment variables without validation. REDMINE_API_KEY and REDMINE_HOST are read from process.env at server startup but never validated. If missing, the Redmine client is instantiated with undefined values, likely causing all calls to fail with cryptic errors.
No input validation. The tool accepts limit and offset as numbers but does not validate ranges, negative values, or type coercion. An LLM could pass limit=-5 or offset='abc' and trigger a confusing API error.
Response contains raw, unstructured JSON from Redmine API. The tool dumps the entire API response as a JSON string with no filtering, structuring, or simplification. High token count dilutes signal and risks context window exhaustion. Return only fields the agent needs (issue ID, title, status, assignee, etc.).