MCP endpoint that proxies requests to the built-in EmDash MCP, merging in inbox plugin tools and RSS aggregator plugin tools. Handles authentication via Bearer token and manages tool list caching.
Strong foundation with clear verb-noun naming (list_threads, get_thread, search_messages, mark_read, pin_thread, snooze_thread, mark_done) and comprehensive parameter descriptions. All 7 tools have documented input schemas with proper type constraints (enums, string minLength, integer bounds). Descriptions are well-written (100-200 chars, within baseline 194 avg). However, output schemas are not documented, LLMs cannot predict response structure for downstream chaining. No error handling guidance visible. Missing tool annotations (readOnlyHint/destructiveHint) despite clear risk classification (READ_ONLY vs WRITE). Parameter descriptions are strong but lack dependency hints (e.g., 'Call list_threads first to get threadId').
Get all messages in one thread, sorted chronologically. Returns full message bodies (text + html), sender, recipient, timestamps. Use this after list_threads or search_messages to read a conversation in full.
List threads in the inbox, optionally filtered by status (inbox/snoozed/done). Returns thread summaries (id, latest sender, subject, snippet, unread count, message count, sortAt).
Move a thread out of the inbox into the Done folder. Equivalent to clicking 'Done' in the UI.
Mark every message in a thread as read or unread. Equivalent to opening the thread in the UI (which auto-marks read).
Pin or unpin a thread. Pinned threads float to the top of the inbox regardless of date.
Plain-text search across message subject and body. Returns matching thread summaries with the matched message highlighted. Useful for finding conversations by topic when you don't know the thread ID.
Output schemas not documented. LLMs cannot predict response structure (fields, types, pagination) for downstream tool chaining. E.g., list_threads returns 'thread summaries (id, latest sender, subject, snippet, unread count, message count, sortAt)' in prose but no formal schema.
Tool annotations missing. Tools are classified as READ_ONLY or WRITE in metadata but not exposed via readOnlyHint/destructiveHint/idempotentHint in schema. Agents cannot infer safety/retry semantics without explicit hints.
No error handling guidance. Tool descriptions do not explain failure modes or recovery steps. E.g., get_thread(threadId='invalid'), what error is returned? Should the agent retry, call search_messages, or ask the user?
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | <=2025-11-25 | v2 |
Snooze a thread until the specified ISO 8601 timestamp. Snoozed threads disappear from the inbox and reappear at the wake time.
Missing dependency hints in parameter descriptions. E.g., get_thread requires threadId but does not hint 'Call list_threads or search_messages first to obtain threadId.' This forces agents to discover the dependency via trial-and-error.
snooze_thread 'until' parameter uses ISO 8601 format but lacks guidance on timezone handling. Should the agent pass UTC? Local time? Current time + offset? Ambiguity invites incorrect timestamps.