io.github.MkTurner74/botverse

Official registry Productivity

Transcode/conform/splice video, convert docs, transcribe audio for AI agents. Free $1 trial.

What can you do with io.github.MkTurner74/botverse?

  • Video transcoding — MP4 (H.264), WebM (VP9), ProRes 422, GIF, MP3 extraction · $0.25/job
  • Video/audio conform — mux a separate video and audio source into one output, reconciling frame-rate and duration mismatches (including true pulldown-style speed conforms, e.g. 24→25fps) · $0.30/job
  • Conform splice — concatenate video and/or still-image segments into one output: head slates/bumpers, tail slates/end cards, mid-roll inserts/cutaways, or joining clips together · from $0.30/job
  • Document conversion — Markdown ↔ DOCX ↔ PDF ↔ HTML ↔ XLSX · $0.05/file
  • Transcription — speaker-labelled transcripts (diarization + AI speaker naming) → txt/srt/vtt/docx/pdf · ~$3/hour
  • Workflows — chain multiple Botverse operations (transcode, convert, transcribe, conform, splice) into a single multi-step job with automatic dependency ordering and parallel branches, instead of orchestrating each call yourself · convert-only steps run ...

Install

Add this to your client's MCP configuration (Claude Desktop: claude_desktop_config.json; Cursor: .cursor/mcp.json; VS Code and others use the same "mcpServers" shape).

NPM package botverse-mcp
Claude Code
claude mcp add botverse -e 'BOTVERSE_API_KEY=<your botverse_api_key>' -e 'BOTVERSE_CONNECTOR_URL=<your botverse_connector_url>' -- npx -y botverse-mcp@1.5.1
JSON configuration (Claude Desktop, Cursor, VS Code)
{
    "mcpServers": {
        "botverse": {
            "command": "npx",
            "args": [
                "-y",
                "botverse-mcp@1.5.1"
            ],
            "env": {
                "BOTVERSE_API_KEY": "<your botverse_api_key>",
                "BOTVERSE_CONNECTOR_URL": "<your botverse_connector_url>"
            }
        }
    }
}

Environment variables

  • BOTVERSE_API_KEY secret Your Botverse API key (bv_live_...). Generate at https://botverse.cloud/dashboard/api-keys
  • BOTVERSE_CONNECTOR_URL secret Your Botverse connector URL (https://botverse.cloud/mcp?token=bv_sess_...). Alternative to API key — recommended for claude.ai and browser clients.

Documentation

From the project's README on GitHub, refreshed weekly. Links open on GitHub.

Contents
  1. botverse-mcp
  2. What it does
  3. Setup
  4. Usage
  5. Command line (botverse)
  6. Tools (MCP)
  7. Workflows
  8. Pricing
  9. Links

botverse-mcp

MCP server and command-line tool for Botverse — video transcoding and document conversion for AI agents and the humans who configure them.

What it does

  • Video transcoding — MP4 (H.264), WebM (VP9), ProRes 422, GIF, MP3 extraction · $0.25/job
  • Video/audio conform — mux a separate video and audio source into one output, reconciling frame-rate and duration mismatches (including true pulldown-style speed conforms, e.g. 24→25fps) · $0.30/job
  • Conform splice — concatenate video and/or still-image segments into one output: head slates/bumpers, tail slates/end cards, mid-roll inserts/cutaways, or joining clips together · from $0.30/job
  • Document conversion — Markdown ↔ DOCX ↔ PDF ↔ HTML ↔ XLSX · $0.05/file
  • Transcription — speaker-labelled transcripts (diarization + AI speaker naming) → txt/srt/vtt/docx/pdf · ~$3/hour
  • Workflows — chain multiple Botverse operations (transcode, convert, transcribe, conform, splice) into a single multi-step job with automatic dependency ordering and parallel branches, instead of orchestrating each call yourself · convert-only steps run on wallet balance ($0.05/step); transcode/transcribe steps bill by source duration and require auto-refill

Two ways to use it: an MCP server for your AI agents, and a botverse CLI for the shell — evaluation, CI/CD, cron, scripts, and local coding agents. No AWS. No FFmpeg. No infrastructure.

Setup

  1. Sign up at botverse.cloud — free to try: $1 credit on signup, no card required. A card + 2FA are only needed at your first top-up ($5 min). No monthly fees.
  2. Get an API key or connector URL from your dashboard
  3. Add to your MCP client config

Usage

Claude Desktop / Cursor / Windsurf
{
  "mcpServers": {
    "botverse": {
      "command": "npx",
      "args": ["-y", "botverse-mcp"],
      "env": {
        "BOTVERSE_API_KEY": "bv_live_..."
      }
    }
  }
}

Or with a connector URL (recommended for claude.ai):

{
  "mcpServers": {
    "botverse": {
      "command": "npx",
      "args": ["-y", "botverse-mcp"],
      "env": {
        "BOTVERSE_CONNECTOR_URL": "https://botverse.cloud/mcp?token=bv_sess_..."
      }
    }
  }
}

Command line (botverse)

The same package ships a botverse CLI for the shell — it reads files from disk and streams them straight to the API (no content goes through an LLM), so it's the fast path for evaluation, automation, and local coding agents.

export BOTVERSE_API_KEY=bv_live_…        # or BOTVERSE_CONNECTOR_URL=…?token=bv_sess_…

npx botverse convert report.md --to pdf
npx botverse convert *.md --to docx,pdf -o ./out
npx botverse transcode clip.mov --to mp4 -o ./out
npx botverse transcribe call.mp4 --to docx --attendees "Sarah Chen,Mike Torres"
npx botverse conform video.mp4 audio.wav --to mp4 --method speed_conform --target-framerate 25
npx botverse balance

Each job uploads → polls → downloads the finished file to -o (default: current dir). Globs and multiple --to formats run as a batch.

Sandbox note: the CLI needs outbound network to botverse.cloud and Botverse's file storage host, so it does not run inside sandboxed agent environments (claude.ai / Claude Desktop), whose egress is allowlisted. There, use the MCP tools (e.g. convert_content) instead.

Tools (MCP)

ToolDescription
get_upload_urlGet a presigned URL to upload a file for use with the _media/_file/_video (object-key-based) tools
transcode_from_urlTranscode video from a public URL
transcode_videoTranscode an uploaded video file
conform_from_urlMux a separate video + audio source (public URLs) into one output
conform_mediaMux an uploaded video + audio source into one output
conform_splice_from_urlConcatenate video/image segments (public URLs) — slates, bumpers, end cards, inserts, joined clips
conform_splice_mediaConcatenate uploaded video/image segments — same use cases
convert_contentConvert document content inline (up to 4 MB; sandbox-safe)
convert_from_urlConvert a document from a public URL
convert_fileConvert an uploaded document
submit_workflowSubmit a multi-step BWDL workflow chaining several of the above tools together, with dependency ordering and parallel branches
get_workflow_statusPoll a workflow (and its per-step status) until it reaches a terminal state
cancel_workflowCancel an in-progress workflow; already-completed steps are billed, the rest are not
get_job_statusPoll a job until complete
get_download_urlGet the signed download URL
get_wallet_balanceCheck wallet balance

Workflows

submit_workflow chains multiple Botverse tools (transcode, convert, transcribe, conform, splice) into one server-side job — e.g. transcribe → clip extraction → delivery — instead of your agent making each call and polling each one individually.

  • Workflow definitions use BWDL (Botverse Workflow Definition Language): a JSON object with a workflow_id and a steps array. Each step has an id, a tool (any MCP tool name), an inputs object, and an optional depends_on array of prior step ids.
  • Steps run in dependency order; steps that share the same depends_on run in parallel.
  • A later step can reference an earlier step's output with "$.steps.<step_id>.output_key", and a workflow-level parameter with "$.params.<name>" — no other template syntax is supported.
  • Submit with submit_workflow, poll with get_workflow_status every 5–10 seconds until the status is COMPLETED, FAILED, PARTIALLY_FAILED, or CANCELLED, and stop early with cancel_workflow if needed (you're only billed for steps that already completed).
  • Billing: convert-only workflows run on wallet balance at $0.05/step. Workflows containing a transcode or transcribe step require auto-refill to be enabled (billing scales with source duration).

Example — a two-step conversion chain (URL → DOCX → PDF):

{
  "workflow_id": "md-to-pdf-via-docx",
  "steps": [
    {
      "id": "to_docx",
      "tool": "convert_from_url",
      "inputs": {
        "source_url": "https://example.com/report.md",
        "output_format": "docx"
      }
    },
    {
      "id": "to_pdf",
      "tool": "convert_file",
      "depends_on": ["to_docx"],
      "inputs": {
        "object_key": "$.steps.to_docx.output_key",
        "output_format": "pdf"
      }
    }
  ]
}

Pricing

  • Video transcode (≤5 min): $0.25/job
  • Video overage: +$0.08/min
  • ProRes 422: $0.35/source-min ($0.50 min)
  • Document conversion: $0.05/file

Credits never expire. Full pricing →

Frequently asked questions

What is io.github.MkTurner74/botverse?

Transcode/conform/splice video, convert docs, transcribe audio for AI agents. Free $1 trial.

How do I install io.github.MkTurner74/botverse?

It runs on your machine. In Claude Code: claude mcp add botverse -e 'BOTVERSE_API_KEY=<your botverse_api_key>' -e 'BOTVERSE_CONNECTOR_URL=<your botverse_connector_url>' -- npx -y botverse-mcp@1.5.1. For other clients, paste the JSON configuration from this page into their MCP settings.

Does io.github.MkTurner74/botverse need an API key?

Yes. It expects BOTVERSE_API_KEY, BOTVERSE_CONNECTOR_URL, which you set in its configuration.

Is io.github.MkTurner74/botverse open source?

Yes. Its code is on GitHub under the MIT license.

Who maintains io.github.MkTurner74/botverse?

It is published by MkTurner74. AI.tn lists it but does not maintain or endorse it.

From the official MCP Registry. Listing is not an endorsement: review a server's code and permissions before giving it access to your data.