Docsie MCP server · Video analysis

Analyze a video from Claude, ChatGPT or your IDE.

Give your AI assistant a video and a question. Docsie Video Analysis watches the frames and the audio and sends back every finding with its time, the frame and the exact on-screen text, right in the chat.

Connect your AI tool →Book a demo

Runs the same Video Analysis app as docsie.io/video-analysis. Uses AI credits.

https://app.docsie.io/mcpWorks in Claude, ChatGPT, Claude Code, Cursor, Codex and any MCP client.

Every error in a Windows Update screen recording, quoted with its times — 01
Every error, quoted down to the code
A Video Analysis result on a 6:57 screen recording: "Error 0x80070643" each time it appears, with what the narrator says about it. The same app runs from your AI assistant over MCP.

What you can ask

Ask a video what you need to know.

Your assistant runs the Video Analysis app with one of six starting points (moments, a question, a count, on-screen text, errors or hazards) and reads the findings back to you.

Claude

“Find every time the error dialog appears in this screen recording and quote the error code.”

app_run · video-analysis
app_run_status
app_run_result

Each appearance with its timestamp, the frame and the error text copied as shown.

ChatGPT

“Count the deliveries at the loading dock in this camera export and give me the times.”

app_run · count
app_run_result

A count with every sighting listed by time, so you can check each one on its frame.

Cursor

“Read every on-screen message in this bug-report recording, word for word, with who sent it.”

app_run · screen_text
app_run_result

The messages in order, copied as they appear, each linked to its moment in the video.

Watch it work

See a video analyzed, start to finish.

Walkthroughs recorded in the Docsie app. Over MCP, your assistant runs the same app and gets the same results back in the chat.

**How to analyze a video with AI: count every animal in trail-camera footage**Docsie Video Analysis counts every animal in almost 12 minutes of trail-camera footage, by species, with the date and time read off the camera's own clock. Then you click a time to play that moment, search the findings and export the result.Footage: trail-camera video by Forest Moments (CC BY) and goldfinch feeder video by Garden Birds Galore (CC BY). Recorded in Docsie's test environment. No customer footage.

**Extract a chat thread from a screen recording: every message, sender and time**Docsie reads a screen recording of a message thread and writes out every message word for word, with who sent it and the moment it appears. Each message links back to its frame, so a person checks it before it goes into a report.Recording: a synthetic screen recording generated for testing (a made-up maintenance chat). No real people, and no case or customer footage.

**Extract text from video with AI: an ROV telemetry overlay, word for word**The telemetry overlay of an underwater ROV video (date, time, depth, pitch, roll and water temperature) is read into 34 timestamped findings that you check against the video, search, and export as PDF, Word or CSV.Footage: "Deep-sea corals, rockfish, and net in Gull Island Reserve, Channel Islands" by Peter Etnoyer (CC BY). No customer footage.

**Extract frames from a video with AI: one still per phase**Instead of thousands of near-identical frames, Docsie returns one clean still for each phase of NASA's Artemis II mission animation, each with a timestamp and an alt-text caption, then exports the report.Footage: "Artemis II to the Moon: Launch to Splashdown", NASA, public domain. Use of NASA imagery does not imply NASA endorsement of Docsie.

Real results

Real results from Docsie Video Analysis.

Results from the Video Analysis app on sample footage in Docsie's test environment. Over MCP, the same findings come back to your assistant as text with timestamps and frame links.

Safety hazards in a training video, summarised with their times and frames — 02
Each hazard, and what makes it one
Every safety hazard in a 2:56 training video, with the narrator's safety lines quoted alongside and a summary grouped by hazard.

A chat thread from a screen recording, every message with its sender and time — 03
A chat thread, with who sent each message
All 21 messages from a screen recording, in order and word for word, each linked to the moment it's on screen.

Animals on a trail camera counted by species, with the camera's own date and time — 04
Count by type, with the camera's own clock
25 sightings by species in an 11:49 trail-camera video, each with the date and time read off the image.

How it works

How video analysis works over MCP.

Point it at the video

Paste a YouTube, Loom or Vimeo link or a direct video URL, or upload the file to Docsie first.

Say what to look for

Ask in plain language, or name a starting point such as errors, hazards, a count or on-screen text.

Check the findings

Each finding comes back with its time and a link to the frame, so a person can confirm it.

Permissions and limits

What to tick, and what it costs.

Tick Workspace create and edit

Running an app needs the Workspace create and edit pack, which is off by default on Docsie's sign-in screen. Reading results only needs search and read.

AI credits

Every analysis uses your organization's AI credits, the same as running Video Analysis in Docsie.

Up to 40 minutes per video

For longer recordings, split them into parts and analyze each one.

About Docsie Video Analysis →

FAQ

Common questions

Can Claude or ChatGPT analyze a video?

On their own they mostly work from a transcript or a few sampled frames. Connected to the Docsie MCP server, they can run Docsie Video Analysis, which checks frames across the whole video and returns each finding with its timestamp and frame.

Which video sources work?

YouTube, Loom and Vimeo links, direct video file URLs, and files uploaded to your Docsie workspace.

What can I ask for?

Moments that match a description, answers to a question, counts, on-screen text, errors and safety hazards. You can always phrase your own question.

Can it identify people?

No. It describes what is visible, such as clothing, vehicles and actions. It does not do face recognition.

Does it cost extra?

The MCP server is on every Docsie plan. Each analysis uses AI credits, the same as in the app.

Docsie MCP server

Ask your videos from the tools you already use.

Connect Claude, ChatGPT or Cursor to Docsie and run Video Analysis without leaving the chat.