A monorepo containing multiple MCP servers for different services: Kan (kanban task management), Outline (wiki), and Radicale (CalDAV/CardDAV)
Scoring was not performed
Destructive tools (kan_delete_board, kan_delete_list, kan_delete_card) lack confirmation step or dry-run support. Descriptions do not warn of irreversibility or guide recovery.
No output schema documentation. Tool implementations return raw JSON.stringify(data) with no guidance on response structure, fields, or which fields to use in downstream calls. LLMs cannot plan multi-step chains without this information.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 16 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 48 | - | v1 |
Descriptions are too brief (average ~60 chars for tools, well below optimal 50-200 range). Most read as API docstrings rather than LLM-optimized prompts. Missing context: WHEN to use this tool, what problem it solves, what to do with the result.
Error handling is minimal. Generic 'kan.bn API error <status>: <text>' message provides no recovery guidance. No indication which errors are retryable, which are user-fixable, which are fatal.
No idempotency guarantees documented. Tools like kan_create_board, kan_create_list, kan_create_card will create duplicates if called twice with identical parameters. Agents retry on ambiguous failures, risk of double-creation.
No pagination or result-limiting documented for list tools (kan_list_workspaces, kan_list_boards). If a workspace has 1000 boards, raw response will be returned without truncation, exhausting LLM context and wasting tokens.
Parameter descriptions lack specificity. E.g., 'Board name' for kan_create_board name parameter does not state length limits, character restrictions, or uniqueness rules. LLMs may pass invalid values.
No permission scopes declared. Tools allow destructive operations (delete) and write operations (create, update) but do not state what permissions are required or what the agent should validate before invoking.
kan_search query parameter lacks guidance on format. Does it support regex, quotes, operators? If not, saying so prevents LLM from attempting advanced queries.