mac-voice-mcp

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations declare the safety profile (destructive=false, idempotent=true), and the description adds substantial context beyond them: a greeting is spoken on start, the rules of the call are returned, and hangup can be triggered by the user ('hang up', 'bye Claude') or by long silence. Remaining gap: it does not say what happens if start is invoked while a call is already active, which is exactly where idempotentHint=true would matter.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Is the description appropriately sized, front-loaded, and free of redundancy?

Front-loaded with the core definition, then actions, then in-call routing — good ordering with no filler sentences. The 'like a phone call, until they hang up' clause is slightly verbose but does real semantic work by conveying the open-mic model, so the cost is minor.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

One required enum parameter, annotations covering safety, no output schema to explain — the description covers the full lifecycle (start, in-call, implicit and explicit end) at the right level of detail for a tool this small. Nothing an agent needs to invoke it correctly is missing.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

With one enum parameter at 100% schema coverage, the baseline is 3; the description earns above that by adding outcomes the schema omits — start returns the rules of the call, and end may occur without an explicit call. The schema's own enum text ('take the mic and greet the user') is thinner than the description's explanation.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Does the description clearly state what the tool does and how it differs from similar tools?

States a specific verb and resource pair ('Start or end a call') and immediately defines the resource in operational terms (hands-free, mic stays open, user talks freely). An agent can distinguish this from 'say' and 'listen' without opening either schema, since the description frames those as in-call primitives.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Does the description explain when to use this tool, when not to, or what alternatives exist?

Gives explicit when-to-use for each action ('action start takes the speaker and microphone', 'action end hangs up') and routes in-call behavior to the correct siblings: 'During the call, speak with say and hear the user with listen.' It also lists the implicit conditions under which the call ends, so the agent knows ending is not solely its job.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Beyond the annotations, it discloses the key behaviors: it returns immediately if data is pending, otherwise blocks up to wait_seconds (default and maximum 40), emits the literal sentinel string '(still on the call, nothing said)', and reaches a terminal state after the call ends. This explains the non-idempotent, state-advancing nature implied by readOnlyHint=false and idempotentHint=false.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Is the description appropriately sized, front-loaded, and free of redundancy?

Three short sentences, front-loaded with the situational trigger ('During a call'), followed by behavior, sentinel interpretation, and the terminal rule. No filler, and each sentence carries distinct operational information.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

With one optional parameter and no output schema, the description covers everything an agent needs: return semantics, blocking/timeout behavior, the exact sentinel string, and when to stop calling. Return values are described in prose, compensating for the absent output schema.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100% and the single parameter is fully documented there, so the baseline is 3. The description adds meaning beyond the schema by explaining the conditional behavior: the wait only happens if nothing is already waiting, making wait_seconds a maximum blocking time rather than a fixed sleep.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Does the description clearly state what the tool does and how it differs from similar tools?

The description opens with a precise verb+resource and scope: 'During a call: return what the user has said since you last heard from them.' The delta semantics ('since you last heard from them') distinguish it from a full transcript read and implicitly from the sibling 'say'/'speak_and_listen' tools, which produce speech rather than consume it.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Does the description explain when to use this tool, when not to, or what alternatives exist?

It gives explicit lifecycle guidance: call it again to keep waiting when the sentinel '(still on the call, nothing said)' is returned, and stop calling once the call-ended signal appears. What's missing is an explicit statement of when to prefer this over 'speak_and_listen' or 'say', so it stops short of full alternative routing.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Beyond the annotations, the description discloses that playback is non-blocking, that the result includes anything the user said meanwhile, and that tool output does not render on screen. These are important behavioral traits not represented in the annotations.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Is the description appropriately sized, front-loaded, and free of redundancy?

Purpose and routing are front-loaded, and the bullet structure is scannable. However, the description runs long for a one-parameter tool and repeats the on-screen reply warning across two bullets, so it is useful but not maximally concise.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

It covers when to use the tool, the alternative, playback and return behavior, speech composition constraints, and the on-screen reply caveat. With no output schema, it adequately explains what the result includes and what the agent must do to avoid a silent interaction.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Although the schema fully covers the single text parameter, the description adds concrete semantics: 1–3 short sentences, under about 40 words, plain words only, no markdown or URLs, describe code rather than reading it, and ask one question at a time. This is substantial meaning beyond the schema's 'plain, conversational text.'

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Does the description clearly state what the tool does and how it differs from similar tools?

The description states a specific action: speak supplied text during a call and return immediately. It also names the sibling alternative, speak_and_listen, for use outside a call, so an agent can distinguish it from related voice tools.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Does the description explain when to use this tool, when not to, or what alternatives exist?

It explicitly restricts usage to during a call and directs the agent to speak_and_listen otherwise. Additional guidance covers how to compose spoken text, leaving little ambiguity about when and how to invoke it.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Goes well beyond the generic annotations by disclosing the blocking listen cycle (waits for speech to start, stops when the user finishes), the on-device transcript return, the non-blocking behavior of listen:false, the canned '(No speech detected …)' string, and the trailing diagnostics line. It stops short of stating hard limits on audio length or error modes beyond the no-speech case.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Is the description appropriately sized, front-loaded, and free of redundancy?

Purpose is front-loaded in the first sentence and the good/bad example pair earns its space. The block is long, and some lines (e.g. the reminder that Bash stdout is not the on-screen message) drift into coaching on surrounding agent behavior rather than the tool itself, but nearly every sentence carries operational value.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

There is no output schema, and the description covers the return value (on-device transcript, no-speech sentinel, diagnostics line), the conversational continuation contract, and the failure fallback to voice_setup. An agent has everything it needs to invoke this correctly across its main modes.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is already 100%, but the description still adds real semantic value: it defines when listen:false is appropriate, explains why listen_seconds is rarely needed, and gives content-level rules for text_to_speak (plain words only, 1–3 short sentences, describe code instead of reading it, ask one answerable question).

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Does the description clearly state what the tool does and how it differs from similar tools?

The first sentence states a precise verb pair and resource: speak text aloud, then capture the user's spoken reply as a transcript. It also implicitly separates itself from the sibling voice_setup by naming that tool only for the setup-failure case.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Does the description explain when to use this tool, when not to, or what alternatives exist?

Explicit routing rules: keep the loop in voice via repeated calls, pass listen:false to end voice mode or make a one-way announcement, call voice_setup when diagnostics report it is not configured. It also states how to handle 'No speech detected' (ask once more, then fall back to screen text) and that transcripts must never be treated as consent.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already declare readOnlyHint=false, idempotentHint=true, destructiveHint=false and openWorldHint=true, so the safety profile is partly covered. The description still adds real behavior: existing installs are reused, a model found elsewhere on disk is symlinked rather than re-downloaded, brew install runs for missing packages only, and nothing is ever uninstalled. The one notable gap is what the check actually returns beyond the INSTALLING state.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Is the description appropriately sized, front-loaded, and free of redundancy?

The purpose is front-loaded in the first clause and each following sentence covers a distinct concern: reuse of existing installs, the consent gate, kokoro, bubble, and the background INSTALLING state. The main defect is redundancy with the schema — the ~1 GB Kokoro size and 'ask the user first' warnings appear twice — plus an odd line-break inside the first sentence.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a three-flag, open-world setup tool with no output schema, it covers the essentials: prerequisites, consent, side effects, non-destructiveness, and the asynchronous INSTALLING state. What it does not specify is the shape of the check result (how the missing pieces are reported), which is the one thing an agent must parse on the very first call.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100%, so all three booleans are already described in the input schema, including the 'ask the user first' caveat. The description adds workflow meaning ('install=false is the default, call it first') and the post-install consequence that Kokoro is used automatically with the built-in voice as fallback, but it largely restates the schema's own parameter text, so a baseline 3 is right.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Does the description clearly state what the tool does and how it differs from similar tools?

The description names a specific verb+resource: checks the machine for the prerequisites speak_and_listen needs (TTS, SoX, whisper.cpp, speech model) and optionally installs what is missing. It explicitly ties itself to the sibling speak_and_listen, so an agent can place it in the tool set without opening a schema.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Does the description explain when to use this tool, when not to, or what alternatives exist?

It gives a hard ordering rule: call with install=false (the default) first, report what is missing, and only then call install=true after the user agrees. It extends the same consent gate to kokoro and bubble, and adds a re-entry rule for the INSTALLING state (wait a minute and call again).

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.