Loading ToolBench
MCP server for Tellie: send text to the silent second screen in the Mac notch.
Tellie MCP has well-structured tool definitions with clear, action-oriented names (update_status, flash_status, clear_notch, send_to_tellie, read_notch, read_log). All 6 tools have descriptions (avg ~150 chars, within 10-1024 baseline). Input schemas are present with typed parameters and descriptions. However, output schemas are not documented, LLMs cannot predict return structure. Error handling guidance is absent. Parameter descriptions are good but lack format constraints (e.g., icon field accepts 'SF Symbol name or emoji' but no validation rules). No tool annotations (readOnlyHint/destructiveHint) despite clear read vs. write semantics.
Remove status from the notch. With `source`, removes just that source's line from the multi-source roster (use this to sign off when you're done). Without `source`, clears everything.
Show a brief status in the user's notch that auto-clears after a few seconds. Use for one-off pings / milestones (e.g. 'PR opened', 'deploy finished'). Like update_status but transient.
Read the notch history (Tellie Pro records it). Lets you check what happened recently, react to past events, or audit the notch activity. Reads local log files; no effect on the notch. Always available (free).
Read what's currently on the user's notch: the live status roster (each source's current line) and any parked send. Lets you check whether another agent is already reporting, avoid duplicate work, or react to the current state. Reads a local snapshot file; no effect on the notch. Always available (free).
Display text on Tellie's silent second screen in the Mac notch. Tellie parks the text quietly in the notch (it does not steal focus or take over the screen); the user clicks the notch to read it as a teleprompter. Use this to hand the user a script, talking points, or a note to read aloud or refer to while they keep working.
Output schemas not documented. LLMs cannot predict return structure for any tool (update_status, flash_status, send_to_tellie, read_notch, read_log). Agents must guess what fields are available, risking failed downstream reasoning.
No tool annotations (readOnlyHint/destructiveHint/idempotentHint). read_notch and read_log are clearly read-only; update_status, flash_status, send_to_tellie, clear_notch are destructive writes. Annotations enable safer agent planning.
Icon parameter accepts 'SF Symbol name or emoji' but no validation rules or enum. LLMs may pass invalid symbols. Constraint should be formalized (e.g., regex pattern or enum of common symbols).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2026-07-28+ | v2 |
Set or replace a glanceable status line in the user's Mac notch via Tellie. Use it to show what you're doing right now (a build step, a test run, tokens/cost used, a timer, a milestone, anything). It REPLACES the previous line from the same source and never steals focus or interrupts the user. Set attention=true when you need the user to look up (you finished, or you're blocked waiting on them).
No error handling guidance. Tools lack recovery hints (e.g., 'If Tellie is not running, install from tellieapp.com'). Agents cannot self-correct on failure.
read_log sinceHours parameter lacks bounds. No min/max specified. LLMs could pass 0, negative, or absurdly large values (e.g., 999999 hours). Should constrain to 1 - 8760 (1 year).