Use the normalized transcript field when the next step needs clean text for search, notes, analysis, or generation.
Remote MCP ยท OAuth
YouTube transcripts in Codex.
Give Codex direct access to timestamped YouTube transcripts without copying text between browser tabs, while preserving source timestamps for review and citation.
Setup
Connect in three steps.
Connection method: https://threadmap.online/mcp
- 01
Open MCP settings
Add a remote MCP server in Codex and use the Threadmap server URL.
- 02
Authenticate
Choose Authenticate. Threadmap opens a secure OAuth approval page; no API key needs to be pasted into Codex.
- 03
Ask about a video
Share a public YouTube URL and ask for a summary, key points, quotes, facts, timestamps, or the complete transcript.
Working example
MCP server URL
Codex calls Threadmap only when a YouTube question needs the video's spoken content. The transcript remains the source, while Codex handles summarization, comparison, extraction, or writing in the conversation with traceable timestamp references.
https://threadmap.online/mcpWhat comes back
Text for humans, segments for software.
Use segment start times and duration values for citations, chapter links, clips, interfaces, or traceable AI answers.
Language, video duration, transcript mode, cache state, and request IDs make workflows observable and easier to debug.
Production checklist
Make the Codex workflow reliable.
Test with a short public video before adding loops or downstream AI calls. Keep the source URL beside every stored transcript and expose request errors clearly instead of silently skipping videos.
Protect credentials
Keep the Threadmap credential inside Codex or a server-side secret store. Never place a live key in a public webpage, shared document, or client-side script.
Handle failures
Branch on HTTP status and error code. Private videos, unavailable media, rate limits, and insufficient minutes require different follow-up actions.
Monitor balance
Read remaining-minute headers after successful requests or call the usage endpoint before a large batch so the workflow can pause cleanly.
Preserve evidence
Store the video ID and timestamped segments with transformed output. That lets reviewers trace summaries, quotations, and extracted facts back to the source.
Integration questions
Operational details
Which transcript mode should I use?
Use captions for the fastest existing subtitles, auto when you want a supported audio fallback, or transcribe when you explicitly need a fresh audio transcript.
How should Codex retry?
Retry temporary rate-limit and service errors with a delay. Do not repeatedly retry private, unavailable, invalid, or over-limit videos without changing the input.
How is usage measured?
Successful transcript work is billed by video duration rounded up to a whole minute. Purchased credits do not expire, and failed results are not charged.
Can I process many videos?
Yes, but use controlled concurrency, inspect rate-limit headers, and preserve each request ID. For a curated YouTube playlist, the private batch importer may be simpler.
Good workflow design
Keep the source transcript before transforming it.
Save the original transcript and video URL before asking another service to summarize, translate, classify, or rewrite it. That preserves a traceable source, makes retries cheaper, and lets people return to the exact timestamp behind a claim. Respect the video creator's rights when storing, sharing, or republishing transcript content.
Need another connection?
Use the same API from any HTTP-capable tool.
See every supported workflow or open the complete endpoint reference.