Audio and Video Tools for AI Agents: Claude Code, Codex, Cursor, OpenCode, and More

When you ask a coding agent to edit an audio or video file, it usually reaches for whatever command-line tool or Python package is installed on the machine. That works for simple mechanical edits, and hits a wall as soon as the task needs speech or music models.

Combining local scripts with Audo AI's MCP server or API lets the agent handle both:

• Neural voice clean-up and echo removal (remove_noise, enhance_voice). Standard command-line spectral gates (like afftdn or arnndn) muffle speech when background noise changes. Audo AI's remove_noise strips wind, traffic, café chatter, and music while keeping the speaker's real voice, and enhance_voice removes hollow room echo and restores muffled laptop or phone mics ($0.05 a minute, first minute free).
• Stereo stem splitting for songs and videos (split_audio). Separates a song into vocals and instrumental (or 4 stems: vocals, drums, bass, and other), or splits a video's soundtrack into dialogue, background music, and sound effects while copying the original video picture untouched ($0.08 a minute, first 45 seconds free).
• Speaker-labeled transcription and script alignment (transcribe_audio, align_script). Transcribes recordings up to 10 hours in 60 languages with labels for up to 8 speakers and exact start and end times for every word, returning TXT, SRT, VTT, and JSON in one run ($0.03 a minute, first 2 minutes free).
• Filler words, pauses, speaker leveling, and LUFS (remove_fillers, remove_silence, level_speakers, analyze_audio). Cuts ums, uhs, and repeated phrases in the quiet between words ($0.03 a minute), shortens dead air ($0.01 a minute), balances loud hosts and quiet guests ($0.02 a minute), and measures integrated LUFS, true peak, noise floor, and clipping for free ($0.00, 20 files a day).

Every Audo AI surface runs the same 15 tools, the same free tier, and the same per-minute prices:

• 1. Remote HTTP MCP server (https://audo.ai/mcp). In Claude Code, run `claude mcp add --transport http --scope user audo https://audo.ai/mcp`. In Codex, run `codex mcp add audo --url https://audo.ai/mcp`. In Gemini CLI, run `gemini mcp add --transport http --scope user audo https://audo.ai/mcp`. Cursor, VS Code, OpenCode, Zed, Antigravity, ChatGPT, Claude, Cowork, and Raycast connect to the same URL on /setup.
• 2. One-call Free API with curl (no API key or sign-up). Any agent with bash or HTTP access can send `curl -F file=@interview.wav https://audo.ai/api/try/remove-noise -D response.headers -o response.body` (or `/api/try/enhance-voice`, `/api/try/transcribe`, `/api/try/media-info`) right away with no account or key for the free part of each file.
• 3. Keyed REST API (/api/v1/jobs) and OpenAPI spec. For custom agents built with the OpenAI Agents SDK, Claude Agent SDK, Vercel AI SDK, LangGraph, CrewAI, PydanticAI, or n8n (/docs/n8n), pass `Authorization: Bearer $AUDO_API_KEY` to process full files up to 10 hours and 10 GB, with optional auto-refill on /keys.

When you tell Claude Code, Codex, Cursor, or OpenCode to clean up and publish a raw episode or interview in your project folder, the agent can chain Audo's tools in order:

• Step 1: Inspect the recording for free (analyze_audio). The agent calls `analyze_audio` (or `POST /api/try/media-info`) at $0.00 to read integrated LUFS, true peak dBTP, RMS noise floor, clipped samples, and silent timestamp ranges before deciding which clean-up steps the file needs.
• Step 2: Clean the voice and balance the speakers (remove_noise, enhance_voice, level_speakers). The agent runs `remove_noise` if the background is noisy, `enhance_voice` if the room echoes, and `level_speakers` if one person is quieter than the other. On a video file, the picture stays untouched.
• Step 3: Tighten fillers and pauses, then transcribe (remove_fillers, remove_silence, transcribe_audio). The agent removes ums and uhs, shortens long silences, normalizes loudness to -16 LUFS for podcasts (or -14 LUFS for YouTube), and generates speaker-labeled TXT show notes plus SRT and VTT captions next to the original file.

Use each tool for the part of the job it does best:

• Local command-line tools (SoX, MoviePy, pydub). If your agent is already running on a machine with local media binaries and only needs to slice a known timestamp range, concatenate two identical WAV files, or read a file header offline with zero network traffic.
• Audo AI's browser tools and waveform editor. If you want to audition Original and Cleaned audio side by side at the same playhead, drag a strength slider by ear, or visually uncheck individual filler-word cuts on a waveform before exporting.
• Audo AI via MCP or API. Whenever an agent needs speech-aware noise removal, room echo removal, 4-stem song or 3-stem video separation, multi-speaker transcription in 60 languages, speaker leveling, or cloud media processing without installing local weights or GPUs.