Redmine MCP has 4 tools with basic definitions but significant gaps in parameter descriptions, schema documentation, and error handling guidance. Tool names follow verb-noun convention (good). Descriptions exist but are minimal (10-50 chars, below the 194-char baseline). All tools lack documented output schemas. Parameters have types but descriptions are sparse or missing. No error recovery guidance. No idempotency hints. No pagination support despite potential for large result sets. The update_issue_status tool mixes two optional parameters (status_id vs status_name) without documenting mutual exclusivity or validation logic.
Tools (4)
get_issueread onlyauthsource verified68/100
Get specific Redmine issue details
health_checkread onlyauthsource verified72/100
Health check tool to confirm that the server is functioning properly
Output schemas not documented. Tools return strings (formatted text) with no structured field definitions. LLMs cannot reliably extract data or chain to downstream tools.
Parameter descriptions missing or minimal. 'include_details' has a description but issue_id lacks context. 'status_id' and 'status_name' in update_issue_status are not documented as mutually exclusive; LLMs will pass both, causing ambiguity.
No error recovery guidance. Health check returns raw exception strings; no actionable next steps. get_issue has no error documentation, what if issue_id is invalid? LLM receives raw errors with no guidance.
health_checkget_issueupdate_issue_status
HIGH
Recommendations
Document output schemas for all tools. Define structured return types (e.g., {server_name: string, redmine_domain: string, debug_mode: boolean} for server_info; {status: 'healthy'|'unhealthy', message: string, timestamp: ISO8601} for health_check). Enable downstream tool chaining and reliable field extraction.
Expand tool descriptions to 100 - 200 characters. For get_issue, add: 'Retrieves detailed information about a specific issue in Redmine, including status, priority, assignments, and optional attachments. Use this to view issue history before updating status.' For update_issue_status: 'Updates the status of an issue. Provide either status_id (numeric) OR status_name (text), but not both. Supports optional status change notes.'
Add parameter descriptions to all fields. For issue_id: 'Numeric Redmine issue ID (e.g., 42). Required.' For include_details: 'If true, fetch attachments, notes, and relationships; if false, return summary only. Default: true. Set to false for faster responses with high volumes.' For status_id/status_name: 'Provide status_id (integer) OR status_name (string), not both. If unsure of status_id, call list_issue_statuses() first to discover valid options.'
Implement pagination for get_issue when include_details=true. Add optional limit (default 10, max 100) and offset parameters for attachments/journals. Return total_attachments and total_journals counts so LLM knows if results are truncated.
Add error recovery guidance. For health_check: if connection fails, return '{"status": "error", "message": "Cannot reach Redmine at {config.redmine_domain}. Check REDMINE_URL and REDMINE_API_KEY environment variables. Verify network connectivity."}' For get_issue with invalid ID: return '{"status": "not_found", "message": "Issue #999 not found. Call search_issues() to discover valid issue IDs."}' For update_issue_status with invalid status_name: return '{"status": "invalid", "message": "Status \"pending\" is invalid. Valid statuses: open, in_progress, resolved, closed. Call list_issue_statuses() to see full options."}'
No pagination documented. get_issue with include_details=['attachments', 'journals', 'relations'] could return thousands of records, blowing context window. No limit, cursor, or offset parameters.
Tool descriptions too short (10 - 50 chars vs 194-char baseline). 'Get server information and status' lacks context for when to call it. 'Health check tool to confirm that the server is functioning properly' is marginally better but does not explain when an LLM should invoke it vs relying on a background health monitor.
No idempotency or destructive hints. update_issue_status is a write operation but has no annotation or description warning that repeated calls with the same parameters will move the issue multiple times if called twice (e.g., in a retry loop, LLM may not realize the issue is already updated).
Confirmation/dry-run missing for destructive operations. update_issue_status modifies state but offers no confirmation step. Agents can issue status changes without explicit approval, risking unintended state mutations.
Parameter type 'status_name' is a free-form string with no enum. Valid status names are likely constrained by the Redmine server, but LLMs must guess. A 'list_issue_statuses' tool or enum values in the schema would prevent hallucinated invalid statuses.
update_issue_status
Add idempotency and destructiveness hints. For update_issue_status, include a note: 'WARNING: This tool modifies state. Repeated calls with identical parameters will attempt to re-update the issue. For idempotence, check current status via get_issue() before calling this tool a second time.'
Create a companion discovery tool: list_issue_statuses(project_id: optional) → returns [{id: int, name: string}]. This lets LLMs validate status names before calling update_issue_status, preventing invalid input errors.
Add optional 'notes' parameter validation. Document: 'notes: optional string (max 5000 chars) describing the reason for status change. Will be logged as a journal entry on the issue.'
Ensure returned issue data includes all fields needed for follow-up operations. When get_issue returns issue details, include project_id, assignee_id, status_id so LLM can call update_issue_status or reassign without a lookup. Current response appears to include these; verify they are always present.