Skysay

Publish an agent's voice stack

set_agent_voice_stack

Destructive

Save revision-bound shared voice_setup for workflow publication, or explicitly publish an ordinary voice stack, optional outgoing-only version-pinned audio environment, and selected language profiles. Each locale override may set only exact voice, transcriber, and greeting; every resulting stack and existing credential binding is validated before a frozen publication is created. The environment is never caller/STT input. An engine whose agent-catalog row says has_language_axis false (a voice-to-voice engine) refuses languages, language_profiles and follow_caller with a 400 naming the field and the engine.

Input Schema

Name	Required	Description
`model`	No
`voice`	No
`agent_id`	Yes
`language`	No
`languages`	No	When supplied replaces the selected language set.
`listening`	No	Which languages calls listen for: open (default; the transcriber hears the other party in any language when it can) or agent_language (the call's own language only, slightly more accurate there). Frozen in the publication, so changing it republishes. Refused for an engine without a language axis.
`navigation`	No	Per-agent phone-menu navigation (enabled, max_hold_seconds, max_navigation_seconds), validated exactly like PATCH /v1/agents/{agent_id}/voice-stack. On by default: enabled is true, false (off, kept by every save) or null (follow the platform default, on), and is required; an empty object is refused with navigation_enabled_required. Never refused for the engine or the languages: an agent that cannot navigate yet (GPT-Live) keeps the setting and the voice stack reports navigation.eligible and navigation.ineligible_reason.
`transcriber`	No
`turn_taking`	No	Per-agent turn-taking tuning, validated exactly like PATCH /v1/agents/{agent_id}/voice-stack.
`voice_setup`	No	Shared finite input. Requires expected_voice_setup_revision beside it. Do not mix with transcriber or overlapping turn_taking fields. With a workflow, stages input for workflow publication; ordinary calls keep their published stack. Without a workflow, publishes a compatible cascade selection or saves an independent native-engine confirmation listener. Free-form business writes require the finite listener, with additional transcription usage; conversation input remains engine-owned. GPT-Live Transcribe is qualified for finite English input with platform-managed credentials.
`voice_polish`	No	Voice polish. delivery: how the voice performs its speech (natural, lively, calm, precise); unset keeps provider defaults. acknowledgements: while a slow reply is still being prepared, say one short bounded phrase in the agent's own voice and language, at most once per turn, after delay_ms (default 1200) of silence; off unless enabled. Phrases default to the platform's reviewed list for the agent's language; a language without one stays silent unless phrases are supplied. tool_wait: what the caller hears while a slow tool call runs, after delay_ms (default 1000): mode sound plays a quiet typing bed cut when the tool finishes (any language); mode acknowledge speaks one short phrase in the agent's voice (phrases apply to acknowledge only); off by default.
`first_message`	No
`follow_caller`	No	Automatic multilingual policy, on by default where the transcriber can listen for the whole selected set; send it only to record an explicit choice (false stays in the opening language, true is refused where it cannot follow). Omitted scope means selected. All resolves the exact stack's compatible set at publication; provider candidate limits, or the 32-language limit of an AI agent's selection, can require selected instead.
`system_prompt`	No
`voice_frontend`	No	Conversation engine. Omit or cascade keeps STT/LLM/TTS. gpt_live is GPT-Live-1; transcriber/model/voice become the OpenAI projection (captions stub, backend Luna/Terra, gpt-live-1 voice). Switching back to cascade requires a full STT/LLM/TTS stack. Each engine declares whether it has a language axis as has_language_axis in get_agent_catalog.
`answer_detection`	No	Per-agent answer-detection policy, validated exactly like PATCH /v1/agents/{agent_id}/voice-stack.
`audio_environment`	No	Explicit provider-independent ambience pin. Off requires volume 0; every other catalogue preset requires 1 through 100.
`language_profiles`	No	When supplied replaces sparse overrides; omit to preserve them.
`language_profile_revision`	No
`expected_voice_setup_revision`	No

Schema Changelog

Changes observed during successful MCP inspections.

Changed1 schema field changedOct 31, 2026

changedInput schema / properties / follow_caller / description

Previous value: -"Explicit automatic multilingual policy. Omitted scope means selected. All resolves the exact stack's compatible set at publication; provider candidate limits, or the 32-language limit of an AI agent's selection, can require selected instead."New value: +"Automatic multilingual policy, on by default where the transcriber can listen for the whole selected set; send it only to record an explicit choice (false stays in the opening language, true is refused where it cannot follow). Omitted scope means selected. All resolves the exact stack's compatible set at publication; provider candidate limits, or the 32-language limit of an AI agent's selection, can require selected instead."

Changed1 schema field changedOct 31, 2026

addedInput schema / properties / listening

Added value: +{
+  "additionalProperties": false,
+  "description": "Which languages calls listen for: open (default; the transcriber hears the other party in any language when it can) or agent_language (the call's own language only, slightly more accurate there). Frozen in the publication, so changing it republishes. Refused for an engine without a language axis.",
+  "properties": {
+    "mode": {
+      "enum": [
+        "open",
+        "agent_language"
+      ],
+      "type": "string"
+    }
+  },
+  "type": "object"
+}

Changed1 schema field changedOct 27, 2026

changedInput schema / properties / follow_caller / description

Previous value: -"Explicit automatic multilingual policy. Omitted scope means selected. All resolves the exact stack's compatible set at publication; provider candidate limits can require selected instead."New value: +"Explicit automatic multilingual policy. Omitted scope means selected. All resolves the exact stack's compatible set at publication; provider candidate limits, or the 32-language limit of an AI agent's selection, can require selected instead."

Changed1 schema field changedOct 27, 2026

changedInput schema / properties / navigation / description

Previous value: -"Per-agent phone-menu navigation (enabled, max_hold_seconds, max_navigation_seconds), validated exactly like PATCH /v1/agents/{agent_id}/voice-stack. On by default: enabled is true, false (off, kept by every save) or null (follow the platform default, on), and is required; an empty object is refused with navigation_enabled_required. Never refused for the engine or the languages: an agent that cannot navigate yet (GPT-Live, or only Estonian or Finnish) keeps the setting and the voice stack reports navigation.eligible and navigation.ineligible_reason."New value: +"Per-agent phone-menu navigation (enabled, max_hold_seconds, max_navigation_seconds), validated exactly like PATCH /v1/agents/{agent_id}/voice-stack. On by default: enabled is true, false (off, kept by every save) or null (follow the platform default, on), and is required; an empty object is refused with navigation_enabled_required. Never refused for the engine or the languages: an agent that cannot navigate yet (GPT-Live) keeps the setting and the voice stack reports navigation.eligible and navigation.ineligible_reason."

Changed1 schema field changedOct 26, 2026

changedInput schema / properties / voice_frontend / enum

Previous value: -[
-  "cascade",
-  "gpt_live",
-  "gemini_live"
-]New value: +[
+  "cascade",
+  "gpt_live",
+  "gemini_live",
+  "eleven_realtime"
+]

Changed1 schema field changedOct 23, 2026

changedInput schema / properties / voice_setup / description

Previous value: -"Shared finite input. Requires expected_voice_setup_revision beside it. Do not mix with transcriber or overlapping turn_taking fields. With a workflow, stages input for workflow publication; ordinary calls keep their published stack. Without a workflow, publishes a compatible cascade selection. Native free-form input is engine-owned. GPT-Live Transcribe is workflow-only, English and platform-managed."New value: +"Shared finite input. Requires expected_voice_setup_revision beside it. Do not mix with transcriber or overlapping turn_taking fields. With a workflow, stages input for workflow publication; ordinary calls keep their published stack. Without a workflow, publishes a compatible cascade selection or saves an independent native-engine confirmation listener. Free-form business writes require the finite listener, with additional transcription usage; conversation input remains engine-owned. GPT-Live Transcribe is qualified for finite English input with platform-managed credentials."

Changed2 schema fields changedOct 21, 2026

addedInput schema / properties / expected_voice_setup_revision
```
Added value: +{
+  "minimum": 0,
+  "type": "integer"
+}
```

addedInput schema / properties / voice_setup

Added value: +{
+  "additionalProperties": false,
+  "description": "Shared finite input. Requires expected_voice_setup_revision beside it. Do not mix with transcriber or overlapping turn_taking fields. With a workflow, stages input for workflow publication; ordinary calls keep their published stack. Without a workflow, publishes a compatible cascade selection. Native free-form input is engine-owned. GPT-Live Transcribe is workflow-only, English and platform-managed.",
+  "properties": {
+    "listener": {
+      "anyOf": [
+        {
+          "additionalProperties": false,
+          "oneOf": [
+            {
+              "not": {
+                "required": [
+                  "credential_id"
+                ]
+              },
+              "properties": {
+                "credential_source": {
+                  "enum": [
+                    "platform_managed"
+                  ]
+                }
+              }
+            },
+            {
+              "properties": {
+                "credential_source": {
+                  "enum": [
+                    "customer_byok"
+                  ]
+                }
+              },
+              "required": [
+                "credential_id"
+              ]
+            }
+          ],
+          "properties": {
+            "credential_id": {
+              "maxLength": 128,
+              "type": "string"
+            },
+            "credential_source": {
+              "enum": [
+                "platform_managed",
+                "customer_byok"
+              ],
+              "type": "string"
+            },
+            "languages": {
+              "items": {
+                "maxLength": 64,
+                "type": "string"
+              },
+              "maxItems": 128,
+              "minItems": 1,
+              "type": "array"
+            },
+            "model": {
+              "maxLength": 128,
+              "type": "string"
+            },
+            "registry_name": {
+              "pattern": "^[a-z][a-z0-9_]{0,63}$",
+              "type": "string"
+            }
+          },
+          "required": [
+            "registry_name",
+            "model",
+            "languages",
+            "credential_source"
+          ],
+          "type": "object"
+        },
+        {
+          "type": "null"
+        }
+      ]
+    },
+    "turn_taking": {
+      "additionalProperties": false,
+      "description": "Explicit values override the preset. Workflow bounds: 300–3000 ms silence, 0–1000 ms interruption. Cascade bounds: 100–2000 and 80–2000 respectively.",
+      "properties": {
+        "interruption_sensitivity": {
+          "enum": [
+            "balanced",
+            "low",
+            "high"
+          ],
+          "type": "string"
+        },
+        "min_speech_ms": {
+          "maximum": 2000,
+          "minimum": 0,
+          "type": "integer"
+        },
+        "silence_hold_ms": {
+          "maximum": 3000,
+          "minimum": 100,
+          "type": "integer"
+        }
+      },
+      "type": "object"
+    }
+  },
+  "required": [
+    "listener",
+    "turn_taking"
+  ],
+  "type": "object"
+}

Changed2 schema fields changedOct 19, 2026

changedInput schema / properties / navigation / description

Previous value: -"Per-agent phone-menu navigation (enabled, max_hold_seconds, max_navigation_seconds), validated exactly like PATCH /v1/agents/{agent_id}/voice-stack. For an agent on GPT-Live, enabled true is refused with navigation_engine_not_supported (also when switching a navigating agent to GPT-Live)."New value: +"Per-agent phone-menu navigation (enabled, max_hold_seconds, max_navigation_seconds), validated exactly like PATCH /v1/agents/{agent_id}/voice-stack. On by default: enabled is true, false (off, kept by every save) or null (follow the platform default, on), and is required; an empty object is refused with navigation_enabled_required. Never refused for the engine or the languages: an agent that cannot navigate yet (GPT-Live, or only Estonian or Finnish) keeps the setting and the voice stack reports navigation.eligible and navigation.ineligible_reason."

addedInput schema / properties / navigation / required
```
Added value: +[
+  "enabled"
+]
```

Changed1 schema field changedOct 18, 2026

addedInput schema / properties / navigation

Added value: +{
+  "description": "Per-agent phone-menu navigation (enabled, max_hold_seconds, max_navigation_seconds), validated exactly like PATCH /v1/agents/{agent_id}/voice-stack. For an agent on GPT-Live, enabled true is refused with navigation_engine_not_supported (also when switching a navigating agent to GPT-Live).",
+  "type": "object"
+}

First observedOct 15, 2026

TDQS

A3.5/5.0

Behavior4/5

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

The description adds meaningful behavior beyond the annotations: it says every stack and existing credential binding is validated before a frozen publication is created, that the environment is never caller/STT input, and that engines without a language axis refuse certain fields with a 400 naming the field and engine. This is useful context for a destructive publication operation, though it does not cover all mutation and idempotency semantics.

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

Conciseness3/5

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

The description is front-loaded with the save/publish purpose and contains useful details, but the sentences are dense, jargon-heavy, and sometimes awkward. It avoids obvious waste, yet the phrase 'revision-bound shared voice_setup for workflow publication' and similar wording make it harder to parse than necessary.

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

Completeness3/5

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

Given a complex 20-parameter mutation tool with no output schema and only annotation-level safety information, the description covers the main publication behavior and a key engine restriction. It is still not complete enough about revision requirements, interactions among model/voice/transcriber fields, and lifecycle implications for an agent to call it confidently without consulting the schema deeply.

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

Parameters3/5

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

Schema description coverage is only 55% across 20 parameters, so the description must compensate. It adds important meaning for voice_setup, languages, language_profiles, audio_environment, follow_caller, and engine language-axis restrictions, but many parameters such as model, voice, first_message, system_prompt, answer_detection, and revision fields remain thinly explained or rely entirely on the schema.

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

Purpose4/5

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

The description states a specific action: saving or publishing an agent's voice stack, with a distinction between revision-bound shared voice_setup for workflow publication and explicitly publishing an ordinary voice stack. It is clear what the tool does, though it does not directly contrast itself with nearby tools such as get_agent_voice_stack or preview_voice.

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

Usage Guidelines3/5

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

It gives some usage context, including workflow publication versus ordinary publication, and warns not to mix voice_setup with transcriber or overlapping turn_taking fields. However, it does not explicitly say when to use this instead of get_agent_voice_stack, preview_voice, update_agent, or list_audio_environments.

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

All Skysay tools Inspect

Try in Browser

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Glama MCP Gateway

Add one secure layer between your agents and this server.

Publish an agent's voice stack

Input Schema

Schema Changelog

TDQS

Glama MCP Gateway

Resources