Production-ready MCP server with interactive UI rendered inside AI interfaces
This is a solid B-tier MCP server with well-defined tools across 5 functional domains (board, monitor, poll, quiz, timer). Major strengths: all 21 tools have descriptions (194-char baseline met), proper tool annotations (readOnlyHint, destructiveHint, idempotentHint present), structured Pydantic responses, and clear verb-noun naming conventions. Schemas are present and typed. However, several gaps prevent A-tier: (1) parameter descriptions are sometimes generic or missing detail on constraints; (2) output schemas are not explicitly documented in responses, only inferred from model_dump() calls; (3) error handling is basic (try/except returning success/error dict) without recovery guidance; (4) some tools accept free-form strings where enums would help (direction in move_task, action in timer_action have some constraints but not fully enumerated); (5) composite tools like submit_poll_answer and submit_quiz_answer require session/question IDs that chain from prior calls, but dependencies are not well documented. Tool names are strong across the board (verb-first, clear intent), and descriptions are informative. The server demonstrates good composition discipline, each tool has a single responsibility, and UI tools properly link to resources via meta.ui.resourceUri.
Add a new task to the Todo column. Returns updated board state.
Create a new poll with custom questions and options. Returns a poll_id that can be passed to show_poll.
Create a new quiz with custom questions and answers. Each question must have exactly one correct answer. Returns a quiz_id that can be passed to show_quiz.
Delete a task by ID. Returns updated board state.
Export the full board state as a JSON string. Save this to restore the board after a server restart.
Export poll results as JSON or CSV. Pass format='json' or format='csv'. Pass poll_id to export a specific custom poll (omit for default poll).
Parameter constraints not formalized as enums. 'direction' in move_task and 'action' in timer_action accept string values with documented constraints in descriptions (e.g., 'left' or 'right', 'start'|'pause'|'reset'|'set'|'preset'), but these are not enforced as enum types in the JSON Schema. This invites LLMs to hallucinate invalid values.
Output schemas not explicitly documented. Tools return Pydantic model_dump() responses, but the structure is not documented inline or via a schema field. Downstream tools (e.g., submit_poll_answer) depend on response fields (session_id, question_id) that are not formally defined. This forces LLMs to infer structure from examples.
Stateful tool chains not documented. submit_poll_answer requires session_id, question_id, option_id returned from show_poll and implicit user interaction. The dependency chain and expected flow are not explicitly described in tool descriptions, making it unclear when/how to use them sequentially.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 63 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 48 | 2025-03-26+ | v1 |
Get the top scores leaderboard across all quiz sessions.
Get aggregated poll results across all completed sessions. Pass poll_id to get results for a specific custom poll.
Fetch live CPU, memory, disk usage, and top processes. Call this to refresh monitor data.
Replace the current board state with a previously exported JSON string. WARNING: this overwrites all current tasks.
Move a task left or right between kanban columns. Returns updated board state.
Set alert thresholds for system metrics. An alert is raised when a metric meets or exceeds its threshold. Pass only the thresholds you want to change (in percent, 0-100).
Open the interactive kanban board UI. Returns all tasks grouped by column.
Open the live system monitor UI. Returns current CPU, memory, disk, and process stats.
Open the interactive poll UI inside the chat. Returns the first question and a session ID. Pass poll_id to use a custom poll created with create_poll.
Open the interactive trivia quiz UI. Returns the first question and a game session. Pass quiz_id to use a custom quiz created with create_quiz.
Open the interactive countdown timer UI. Returns current timer state (status, remaining seconds, total seconds).
Submit an answer to the current poll question. Returns the next question or final results.
Submit an answer to the current quiz question. Returns score and next question or final leaderboard.
Control the countdown timer. - action='start': start or resume the timer - action='pause': pause the running timer - action='reset': reset to idle (keeps current duration) - action='set': configure duration in seconds (requires seconds parameter) - action='preset': use a named preset: 'pomodoro' (25m), 'short_break' (5m), 'long_break' (15m)
Update an existing task's title, description, labels, or priority. Only provided fields are changed.
Error handling lacks recovery guidance. move_task, update_task, delete_task, and import_board catch MCPAppError and return {'success': False, 'error': str(e)}, but the error messages themselves are not inspected. The tool descriptions provide no guidance on what to do when a task_id is not found or a move direction is invalid.
Complex nested input schemas with minimal constraint documentation. create_poll and create_quiz accept nested objects (questions with options/answers) but do not specify min/max counts, required vs optional fields within the nested structures, or validation rules (e.g., 'exactly one correct answer per quiz question'). LLMs must infer structure from vague descriptions.
Optional parameter descriptions lack guidance on mutual exclusivity. set_monitor_alerts accepts cpu_threshold, memory_threshold, disk_threshold as separate optional params, but it's unclear whether you must provide at least one or if passing none is valid. Description should state 'At least one threshold must be provided'.
Destructive operations (delete_task, import_board) lack confirmation or dry-run support. Agents can irreversibly delete tasks or overwrite the entire board state without a safeguard. No confirmation_before_execute pattern is present.