korovin-aa97/talkthrough-mcp

Overview

Local MCP server that records screen activity and converts it into structured evidence for coding agents. Produces a transcript, video frames, OCR text, and wall-clock timestamps so agents can inspect what happened without a written bug report.

Best for

Best for
Developers who want to hand coding agents a recorded reproduction instead of a written bug report.

Use cases

• Capture a bug reproduction as screen recording and transcript
• Give a coding agent visual and OCR evidence of an issue
• Generate a timestamped trace of UI steps for debugging

How to use

Install

claude mcp add -s user talkthrough -- uvx --python ">=3.11,<3.14" "talkthrough-mcp[diarization,url]"

Tools exposed

• triage-recording
• spec-from-workshop
• backlog-from-demo
• meeting-actions
• correlate-with-logs
• large-v3-turbo
• TALKTHROUGH_WHISPER_MODEL
• TALKTHROUGH_OCR
• TALKTHROUGH_OCR_LANG
• TALKTHROUGH_OCR_PARAMS
• TALKTHROUGH_DIARIZE
• TALKTHROUGH_DIARIZATION_THRESHOLD
• TALKTHROUGH_DIARIZATION_SEG_MODEL
• TALKTHROUGH_DIARIZATION_EMB_MODEL
• TALKTHROUGH_DIARIZATION_THREADS
• TALKTHROUGH_MAX_SECONDS
• TALKTHROUGH_MAX_FRAMES
• TALKTHROUGH_MAX_DOWNLOAD_BYTES
• TALKTHROUGH_HOME

Tested with

Claude Desktop, Claude Code, Cursor, Windsurf, Cline, Continue, VS Code, ChatGPT

Notes

Local MCP server that records screen activity and converts it into structured evidence for coding agents. Produces a transcript, video frames, OCR text, and wall-clock timestamps so agents can inspect what happened without a written bug report.

24 stars on GitHub. Last updated 2026-09-02. Licensed MIT.

Use cases

• Capture a bug reproduction as screen recording and transcript
• Give a coding agent visual and OCR evidence of an issue
• Generate a timestamped trace of UI steps for debugging

Pros

• Runs locally, keeping recordings private
• Provides multimodal evidence in one MCP interface
• Reduces friction of writing detailed bug reports

Cons

• Requires screen recording, which may be intrusive or blocked in some environments
• Limited to local desktop sessions, not remote or headless workflows
• Young project with few stars, so ecosystem and stability are unproven

Indexed from awesome-mcp-servers-punkpeye and enriched against its public facts.