talkthrough-mcp

# talkthrough-mcp

> Local-first MCP server that turns narrated screen recordings (or any
> video/audio file) into agent-ready structured data: timestamped transcript
> segments, scene-change keyframes, OCR text, and wall-clock anchoring, with
> lazy token-budgeted retrieval tools. No LLM inside the server; recordings
> never leave the machine.

## Docs

- [README](README.md): positioning, quickstart per client, tool reference,
  wall-clock ladder, configuration, CLI
- [Install guide for agents](llms-install.md): machine-followable setup steps
- [Per-engine integrations](integrations/README.md): exact config snippets for
  Claude Code/Desktop, Codex CLI, OpenClaw, Gemini CLI, Cursor, Cline,
  OpenCode, Goose, Copilot CLI, Windsurf, Zed
- [Agent skill](.agents/skills/talkthrough/SKILL.md): how to use the tools well
- [Agent contributor guide](AGENTS.md): working on this repo with a coding agent
- [Troubleshooting](docs/TROUBLESHOOTING.md): first-run downloads, client
  wiring, language/OCR/wall-clock fixes, data removal
- [Design](docs/DESIGN.md): pipeline, manifest schema, token-budget rules
- [Findings contract](examples/output-contract.schema.json): triage output JSON Schema
- [Workflow prompts](examples/prompts/): the 6 server prompts as plain files
- [Contributing](CONTRIBUTING.md): dev setup and gates