MCP server that exposes a self-hosted multi-agent dashboard API (memory, kanban, inter-agent messaging, daily log, bot status) as MCP tools.
Server exposes 15 tools via FastMCP with basic schemas and descriptions. All tools have names starting with action verbs (memory_, kanban_, message_, etc.) and docstrings. However, critical gaps undermine quality: (1) Parameter descriptions are minimal or missing context (e.g., 'card ID' lacks guidance on format or how to obtain it). (2) Output schemas are undocumented, no indication of what fields are returned or their types. (3) Error handling is absent, HTTP errors are raised but not caught or translated into actionable guidance for LLMs. (4) No enum constraints on status/priority/tier parameters, inviting hallucinated values. (5) Secrets (DASHBOARD_TOKEN) are injected server-side (good), but no audit logging or permission gates. (6) Tools like kanban_list perform client-side filtering instead of server-side, wasting bandwidth. (7) No pagination support despite returning lists. Baseline: 549 production tools average 18-char names (this server: 10-20 chars, acceptable); 94% have docstrings (this server: 100%, good); 100% of A+ tools document return types (this server: 0%, critical gap).
List all registered agents with their status (configured/running) and metadata.
Check if a tmux session is alive.
Append to an agent's daily log. Use for: architecture decisions, bug fixes, deploys, lessons learned. Format: "## HH:MM -- Topic\nWhat happened, what was the result"
Archive a kanban card (soft delete).
Add a comment to a kanban card.
List all comments on a kanban card.
No output schemas documented. Tools return dicts/lists but LLMs cannot infer field names, types, or structure. Forces agents to guess what data is available and risks failed downstream tool calls.
Enum constraints missing for status, priority, tier, and mode parameters. Free-form strings invite hallucinated values like 'pending' instead of valid 'planned|in_progress|waiting|done'. LLMs cannot self-correct without explicit enum lists.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 52 | 2026-07-28+ | v2 |
Create a new kanban card.
List kanban cards (client-side filtered).
Move a kanban card to a different status column.
Update an existing kanban card.
Save a new memory.
Search memories — hybrid semantic + keyword by default.
Mark a message as done after processing.
Send an inter-agent message.
List inter-agent messages for an agent.
No error handling or recovery guidance. HTTP errors (raise_for_status) are unhandled; LLMs receive raw exceptions with no actionable next steps. Missing pattern: 'User not found. Try search_users() first.'
Parameter descriptions lack actionable detail. 'card ID' does not explain format, how to obtain it, or whether it's a UUID/integer/slug. 'agent name (default: main)' does not clarify valid agent names or how to discover them.
No pagination support. kanban_list and messages_list return all results client-side filtered, risking context window exhaustion. Missing limit enforcement and next_cursor/offset patterns.