Skip to main content
Glama

generate speech

generate_speech
Destructive

Create one independent billable Studio task. For user-attached or local media in a host with native APIXO attachments, use reference_files with file_bindings. ChatGPT and native Codex connectors use this route. Pass the host-provided attachment or the absolute local path when the connector accepts paths. If the native attachment fails, ask the user to reattach it or supply a usable URL. Do not rehost a failed native attachment without an explicit user request. Put existing media URLs supplied or selected by the user in input. Choose the attachment route by capability. Use reference_files when the client provides native APIXO attachments, including ChatGPT and compatible Codex connectors. Otherwise, for user-selected files accessible to the client, automatically use upload_reference and the APIXO uploader when the client can run Node.js 22+ and reach APIXO. This applies to Claude Code, Cursor, Grok Bot, WorkBuddy, and other clients that meet those conditions. A request to use the selected files in APIXO is sufficient to start the upload; honor the client's execution permissions. Read the model schema and identify each destination URL field before uploading. Submit only confirmed URLs in the user's order. If file access or execution is unavailable, explain the limitation and ask for a usable direct link. Keep upload grants out of replies and show reference URLs only on request. Read list_models.defaults for the default model of this tool, then get_model_schema for the complete schema. generate_task requires model; shortcuts may omit it. Explicit model selects that public task model. Set the documented mode in input; no mode, model fields, defaults or voice choices are inferred or injected. Flat model parameters and legacy reference_images/reference_audio are not accepted. Backend validation is authoritative. Use session_title for a new topic, session_id to continue, and session_id plus base_run_id for a version. Preserve returned model/task_id/session_id/run_id/media_type. Never automatically resubmit or change model/session after rejection or an unconfirmed submission. In user-facing replies, omit technical identifiers by default, including user/profile IDs, task_id, session_id, run_id, base_run_id, voice_id, and file IDs. Retain the exact returned identifiers for tool arguments, task queries, session continuation, version appends, and voice reuse; never redact, regenerate, or guess these references. Show an identifier only when the user explicitly asks for it or a specific troubleshooting step requires that exact value, and show only the reference needed. Prefer plain-language status, model names, result previews/download links, and Studio links built from the returned session_id. This changes reply presentation only; tool result fields remain available for the workflow.

Tool call template

Replace each placeholder with a value matching the input schema. Only required parameters are included; add optional parameters as needed.

{
  "id": 1,
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "arguments": {
      "input": "<input>"
    },
    "name": "generate_speech"
  }
}

Input Schema

NameRequiredDescriptionDefault
inputYesFollow get_model_schema.schema for the original model parameters, including nested objects and arrays. For files bound through reference_files, leave the bound URL fields absent from input. Use direct media URL fields for existing links supplied or selected by the user. Do not mix media URLs with reference_files. Routing, identity and session overrides are not model inputs.
modelNoOmit to use this tool's default from list_models.defaults. Supply a canonical ID or exact slug to select another task model; read its complete schema first.
session_idNoUse a returned or user-provided Studio session ID to add related work to that project. Never guess a session ID.
base_run_idNoTo revise a specific work, pass its returned run_id and the same session_id. Omit to create an independent work in the project.
file_bindingsNoUse paths declared in schema.mcp.attachments. Replace [] with an explicit input array index, e.g. elements[0].image_urls; provide parent objects and their metadata in input. Bind every file once using zero-based file_indices in intended order. Never overwrite direct URLs or guess fields or ordering.
session_titleNoRequired for a new Studio topic: 1-255 characters after trimming. Ignored when session_id is provided.
reference_filesNoUse this field for attached or local media when the host provides native APIXO attachment support. Use the host-provided files or absolute local paths when the connector accepts them, in the intended order. Use file_bindings to select model fields. Do not re-upload a failed native attachment without the user's explicit request. file_id is metadata. Use temporary download URLs only for this call and pass only known file metadata.

Output Schema

NameRequiredDescriptionDefault
errorYes
modelYes
stateYes
run_idNo
task_idYes
progressYes
voice_idYes
created_atYes
media_typeYes
session_idNo
completed_atYes
preview_urlsYes
text_outputsNo
download_urlsYes

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed2 schema fields changed
    • changedInput schema / properties / input / description
      Previous value: -"Follow the complete get_model_schema.schema for model parameters, including nested objects and arrays. Values pass unchanged to backend validation; no plugin field or mode whitelist. Use direct media URLs or reference_files with explicit file_bindings, without mixing them. Routing, identity and session overrides are not model inputs. Metadata does not prove content or access."New value: +"Follow get_model_schema.schema for the original model parameters, including nested objects and arrays. For files bound through reference_files, leave the bound URL fields absent from input. Use direct media URL fields for existing links supplied or selected by the user. Do not mix media URLs with reference_files. Routing, identity and session overrides are not model inputs."
    • changedInput schema / properties / reference_files / description
      Previous value: -"ChatGPT files in original order. file_id is metadata, never a URL or permission proof. Temporary download URLs are used only in this call. Do not invent MIME, byte size, dimensions or duration."New value: +"Use this field for attached or local media when the host provides native APIXO attachment support. Use the host-provided files or absolute local paths when the connector accepts them, in the intended order. Use file_bindings to select model fields. Do not re-upload a failed native attachment without the user's explicit request. file_id is metadata. Use temporary download URLs only for this call and pass only known file metadata."
  2. First observed

TDQS

A3.7/5.0
Behavior5/5

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

Annotations already mark this as destructive and non-read-only, but the description adds substantial operational context: billable task creation, honoring client execution permissions, not rehosting failed attachments without explicit request, treating backend validation as authoritative, not resubmitting after rejection, and strict identifier redaction rules. These disclosures go well beyond 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.

Conciseness2/5

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

The description is very long and repetitive, especially around reference_files and upload_reference routing. It is not front-loaded with what the tool does and buries speech-specific context under generic Studio task workflow instructions.

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?

For a complex tool with seven parameters, nested objects, an output schema, and rich annotations, the description covers attachment routing, sessions, failures, privacy, and delegation to get_model_schema. It remains incomplete because it never explains speech synthesis itself or what text/voice inputs are expected, and it conflates this shortcut with generate_task.

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

Parameters4/5

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

Schema coverage is 100%, so all seven parameters are already documented. The description adds meaningful cross-field routing rules beyond the schema: leave bound URL fields absent from input, do not mix media URLs with reference_files, use file_bindings to select model fields, submit confirmed URLs in the user's order, and omit model to use the default.

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

Purpose3/5

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

The description's opening sentence, 'Create one independent billable Studio task,' describes task creation rather than speech generation and never explicitly says it synthesizes speech. It mentions 'voice choices' and 'voice reuse' in passing, but does not distinguish this tool from generate_task or other generate_* siblings.

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

Usage Guidelines4/5

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

It gives extensive conditional guidance for file routing (reference_files vs upload_reference), session usage (session_title, session_id, base_run_id), and failure handling. However, it does not explicitly say when to choose generate_speech over generate_task or other generation siblings, so the core selection remains inferred.

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

Try in Browser

Glama MCP Gateway

Add one secure layer between your agents and this server.

Resources