$ podcli mcp· 32 tools

Clip podcasts by asking your agent

Podcli adds 32 tools to Claude Code, Codex, Cursor, or any MCP client. Ask in plain words, and your agent transcribes the episode, picks the moments, and renders the clips on your computer.

› Find the best clips in episode 142. Skip anything we already posted.

Claude Code
  1. transcribe_podcastrunning
  2. knowledge_baserunning
  3. clip_historyrunning
  4. suggest_clipsrunning
  5. batch_create_clipsrunning

Eight clips are in podcli-clips/. Two were close to clips you posted in March, so I left them out.

32 tools

Transcribe, score, crop, caption, render, and check what you already posted.

Runs locally

Podcli is a process on your computer. The video never goes to the agent or the model.

Any MCP client

Claude Code, Codex, Claude Desktop, Cursor, and anything else that speaks MCP.

The video stays on your computer

Your agent talks to Podcli through tool calls. Podcli does the heavy work locally, and only text reaches the model.

Your computer

Files

~/shows/long-story

episode-142.mp41.8 GB
podcli-clips/8 files
reads

Local process

Podcli

✓ transcribed 58:12

▸ rendering 6 of 8

tool calls

Your agent

Claude Code

› clip episode 142

calling batch_create_clips

text only

Outside

Your model provider

promptsent
transcript.txt42 KB
episode-142.mp4never

32 tools, each a podcast job

Your agent reads this list and picks the order. You never call a tool by name.

Decide which moments are worth a clip.

  • rank_clip_candidatesReturns a shortlist of sentence-aligned moments so the agent does not have to read the whole transcript.
  • suggest_clipsTakes the agent's picks from the transcript and lines them up as clips.
  • analyze_energyMeasures audio energy to find the loud, lively stretches.
  • modify_clipChanges a clip's timing, title, caption style, or opening hook.
  • toggle_clipSelects or drops a suggested clip before export.
  • knowledge_baseReads your show notes: hosts, voice, audience, banned words.
  • clip_historyLists clips you already made so nothing gets posted twice.
  • record_decisionsSaves this episode's choices, like clip count and caption style.
  • mine_channelFinds clip-worthy moments across a YouTube back catalog without downloading it.

Connected in about five minutes

  1. 01

    Install Podcli

    One download. The first run sets up Whisper, FFmpeg, and the models.

  2. 02

    Connect your agent

    Run one command, or paste one config block.

  3. 03

    Ask for clips

    Name the episode and say what you want back.

Full steps are in the MCP docs.

One command registers Podcli with Claude Code.

Fewer calls, less guessing

A generic FFmpeg server hands your agent scissors. Podcli hands it a finished clip.

A generic FFmpeg MCP

trim · crop · scale · encode

Your agent has to work out:

  • Find the good moments in a 58-minute transcript
  • Work out where each face is, frame by frame
  • Write caption timing and styling by hand
  • Remember what you posted last month
  • Then trim, crop, scale, and encode

Podcli MCP

4 calls

Your agent calls:

  1. 1transcribe_podcastwords with timings
  2. 2clip_historywhat you already posted
  3. 3suggest_clipsthe agent's picks
  4. 4batch_create_clipsframed, captioned, exported

Where Podcli falls short

The MCP server gives an agent the same clipping and multicam tools as the Studio. It is not a DAW.

  • Multicam cuts a full episode to whoever is talking. It is not a DAW: no Overdub, no multitrack mix, no word-level recut.
  • The first run of process, studio, transcribe, or multicam downloads Whisper, FFmpeg, and the models it needs. --help never downloads. podcli mcp does not provision; run podcli setup first.
  • AI clip scoring is optional and runs through your Claude Code or Codex login, or an Anthropic API key. Without those, Podcli scores clips on its own.

Questions

Something missing? support@podcli.com

What is the Podcli MCP server?

Podcli ships as a Model Context Protocol (MCP) server with 32 tools that cover the full podcast-to-clip pipeline: transcription, AI clip scoring against a knowledge base, face-tracked 9:16 cropping, burned-in captions via Remotion, and hardware-accelerated export. It works with any MCP client, including Claude Code, Claude Desktop, Codex, and Cursor, and the agent runs the whole flow.

How do I add Podcli to Claude Code?

Install Podcli (curl -fsSL https://podcli.com/install.sh | sh), then run 'podcli mcp install'. It registers the server with Claude Code and Codex, whichever CLI is on your PATH. For Claude Desktop, add a 'podcli' entry running 'podcli mcp'.

Does the MCP server send my podcast files anywhere?

No. The MCP server runs locally as a process the agent talks to. The video, transcript, and renders all stay on your machine. The only network calls are the optional Claude or Codex requests when you use AI clip scoring.

What can I ask the agent to do?

Anything the 32 tools cover: 'transcribe episode.mp4', 'suggest the top 8 clips', 'change clip 3 to the karaoke caption style', 'render all selected clips in 1080x1920', 'check clip 4 against what we already posted'. The agent picks the right tools and chains them.

How is the MCP server different from the CLI?

The CLI runs a fixed command. The MCP server exposes the same pipeline as separate tools the agent combines on demand. Both ship in the same install, and you can use either or both.

Which clients work with Podcli's MCP server?

Any MCP-compatible client. Tested with Claude Code, Claude Desktop, and Codex. Any agent that speaks MCP can read the tool list and call it.

Give your agent a clipper

Install Podcli, connect your agent, and ask for this week's clips.