Codex CLI Text to Speech: Connect Clonylab and Narrate Scripts

Codex is OpenAI’s coding agent. It runs in your terminal, reads your project and does tasks on request. Like other agents, it works with text and files. Connect a text to speech MCP server and it can also turn a script into speech and save the MP3 beside your code.

Clonylab gives Codex 14 tools: list voices, make a voice-over, watch the job, search your History and fetch the MP3. The tools are the same in every app, and our guide to the text to speech MCP server explains them. This guide is the Codex version, with the config.toml block and one setting that matters for longer scripts.

Codex has an unusual strength for this job. It is happy to run a long chain of steps on its own, such as “read these ten files, write ten scripts, make ten voice-overs and list the results.” That is exactly what the Clonylab tools are good at.

Codex stores MCP servers in ~/.codex/config.toml. A project can add its own .codex/config.toml, but Codex reads that file only in projects you trust. Keep your Clonylab block in the file in your home folder.

Codex’s documentation lists a few more options that are useful here. http_headers sends fixed headers, env_http_headers reads header values from variables, and enabled = false switches a server off without deleting it. You do not need any of them for Clonylab.

A project can also hold a .codex/config.toml with its own servers, for example one that only that project needs. Codex reads that file only in trusted projects, which protects you from a stranger’s repository adding servers without you knowing. For Clonylab, keep the block in your home file. Then every project can use it, and the token stays out of any repository.

You can check what Codex loaded with codex mcp list. It lists the servers you set up. If you ever prefer signing in instead of a token, codex mcp login starts an OAuth login for servers that support it.

Codex gives each tool call 60 seconds by default. Clonylab’s generate_speech waits up to 45 seconds for a result unless the agent asks for a different wait, and it can wait up to 100 seconds. If Codex cuts the call off at 60 seconds, a long script may look like a failure even though the job is still running in your tab.

You have two easy fixes. Ask Codex to check the job again with get_job, or raise the limit in the block:

[mcp_servers.clonylab]
url = "https://clonylab.com/mcp"
bearer_token_env_var = "CLONYLAB_TOKEN"
tool_timeout_sec = 120

Large batches are the most likely place to hit the 60-second limit, so set the longer timeout before you start one.

Start Codex in a project folder. Ask for something small first.

The variable route already helps. The token is not in config.toml, so you can back up or share that file without sharing the token. Keep these habits too.

The Studio stays the best place for one clip that you want to hear and tune. Codex is the better place when voice-overs come out of a larger task. If you also use other agents, our guides for Gemini CLI and Claude Code show the same server in their formats. The pauses page lists the script tags.