The tool split is really a cost decision: `quick_transcribe` at 1 credit is right for a podcast, and wrong for anything with a chart in it. `clip_context` is the one to reach for on long content — it charges flat per call regardless of the video's length, so a 90-minute talk costs the same as a 5-minute one.
A remote MCP server that downloads a video, transcribes it, extracts keyframes, runs OCR and analyses the visuals, then returns one structured document. It exists because transcripts capture what was said and miss what was shown: chart values, UI states, code on screen, dashboard numbers.
- quick_transcribe — timestamped transcript with speaker identification, audio only. 1 credit; typically 5–15 seconds for a 10-minute video
- deep_analyze — the full pipeline: transcript plus keyframe vision plus OCR, returned as summary, key claims with timestamps, visual assets with extracted data, and entities. 5 credits; typically 30–90 seconds
- clip_context — analyse one time range without paying for the whole video. 1 credit in quick mode, 3 in deep
- batch_analyze — up to 10 videos in parallel with cross-video synthesis, priced per video with 10% off at 5 or more
- `focus` on `deep_analyze` tunes the vision prompts toward `technical` or `crypto` without changing the credit cost
- URLs from YouTube, Instagram Reels, Vimeo, Twitter/X, TikTok and direct MP4, WebM or MOV
No local install — connect a client to https://contendeo.app/mcp/ and complete OAuth 2.0 in the browser on first use. New accounts start with 10 free credits and no card. Beyond that: Pro at $12/mo for 100 credits, Power at $39/mo for 500, or pay-as-you-go at $0.15 per credit. Cache hits cost nothing and failed analyses refund automatically. Videos over roughly 20 minutes can approach the client-side request timeout on `deep_analyze` — use `clip_context` on the segments that matter instead.
One command — claude mcp add --transport http contendeo https://contendeo.app/mcp/