VideoDB
VideoDB is a video infrastructure platform for storing, searching, understanding, editing, and generating video, audio, and image content.
Verdict
Common use cases
- Search recorded Zoom calls for action items
- Index training videos for onboarding searches
- Find specific product demo moments by keyword
- Transcribe customer interviews into searchable text
- Organize webinar recordings into queryable collections
Integration
- Vendor
- VideoDB
- Category
- other
- Auth
- API_KEY
- Tools
- 11
- Composio slug
video_db
Tools
- Ask Collection
Queue a grounded question-answer request over a VideoDB collection's Search V2 indexes. This search-family call is metered and requires Search V2 indexes. The returned status only confirms the async request state, not a successful answer. P
- Create Collection
Create a VideoDB collection for organizing and searching media. Collections are private by default.
- Get Async Result
Check one VideoDB async operation by response ID. Returns its current state or terminal provider result without polling repeatedly, and fails when a completed nested provider response reports success:false.
- Get Collection
Get metadata for one VideoDB collection by ID.
- Get Media
Get metadata for one VideoDB video, audio file, or image.
- Ingest Media
Queue ingestion of a video, audio file, or image from an HTTP(S) URL into a VideoDB collection. This non-idempotent call is metered; live billing reported file_upload at 0.09 credits. Queue acceptance is not completion. Pass the returned re
- List Collections
List collections available to the connected VideoDB API key and identify the default collection.
- List Media
List videos, audio files, or images in one VideoDB collection using one normalized media interface. Video results return at most one page and may include an opaque continuation cursor.
- Rename Media
Rename one VideoDB video, audio file, or image.
- Search Collection
Run synchronous semantic search over indexed media in a VideoDB collection. This metered call consumed 0.0015 credits per successful call during live Free_v1 probing, including calls that returned empty fallback results.
- Update Collection
Update the name, description, or visibility of an existing VideoDB collection.
Setup
Setup guide
- 11. Sign up at VideoDB and generate an API key from your account dashboard. 2. In Switchy, open your workspace settings and navigate to the Integrations tab. 3. Click 'Add Integration', search for VideoDB, and select it. 4. Paste your API key into the credential field and click 'Connect'. 5. Switchy will verify the key and confirm the connection. 6. Open any Project, type '@VideoDB list collections' in the chat, and press send. 7. If you see a list of collections (or a prompt to create one), the integration is live. 8. To test ingestion, type '@VideoDB ingest media' followed by a public video URL and the target collection ID. 9. Wait for the async operation to complete, then query the collection with '@VideoDB ask collection' plus your question.
What teammates see: by default, memories from VideoDB are scoped to the Project (PROJECT visibility) - you can mark any memory PRIVATE or share it ORG-wide.
Works well with
Top models
Compatibility data appears once enough Projects have used this MCP together with a given model.
How Switchy teams use it
Starter prompts
Create Media Collection
@VideoDB create a new collection called 'Q1 All-Hands Recordings' so we can index and search every town hall from this quarter.
Ingest Meeting Recording
@VideoDB ingest this meeting recording https://example.com/standup-2024-01-15.mp4 into the 'Team Standups' collection.
Query Across Videos
@VideoDB ask the 'Customer Interviews' collection: 'Which users mentioned pricing concerns or budget constraints?'
List All Collections
@VideoDB list all collections in my account and show me which one is set as default.
Check Ingestion Status
@VideoDB check the status of async operation ID abc123 and tell me if the ingestion completed successfully.
Example outputs
Illustrative - representative of the model's voice and quality, not literal recordings.
@VideoDB list all videos in my 'product-demos' collection and show me their durations and upload dates
Found 8 videos in collection 'product-demos': 1. Q3_Feature_Launch.mp4 (4m 32s, uploaded 2024-01-15) 2. Onboarding_Walkthrough_v2.mp4 (7m 18s, uploaded 2024-01-22) 3. API_Integration_Demo.mp4 (3m 45s, uploaded 2024-02-03) 4. Customer_Testimonial_Acme.mp4 (2m 10s, uploaded 2024-02-14) 5. Dashboard_Tour.mp4 (5m 55s, uploaded 2024-02-28) 6. Mobile_App_Preview.mp4 (3m 20s, uploaded 2024-03-05) 7. Security_Overview.mp4 (6m 40s, uploaded 2024-03-12) 8. Pricing_Explainer.mp4 (4m 15s, uploaded 2024-03-18) This is an illustrative example of the metadata VideoDB returns for each media item in a collection.
This example demonstrates VideoDB's list-media capability. The MCP returns metadata only—not video content itself. Results are paginated; collections with hundreds of videos will require cursor-based continuation. Your API key must have read access to the target collection.
@VideoDB ingest this YouTube video https://youtube.com/watch?v=dQw4w9WgXcQ into my default collection and let me know when it's ready
Queued ingestion of video from https://youtube.com/watch?v=dQw4w9WgXcQ into your default collection. Async operation ID: vdb_ingest_7k3m9p2x Estimated processing time: 2-5 minutes for a typical YouTube video You can check status with '@VideoDB get async result vdb_ingest_7k3m9p2x' or I'll monitor it for you. Once complete, the video will be searchable and available for question-answering workflows. This is an illustrative example; actual operation IDs and timing vary by video length.
Ingestion is metered at ~0.09 credits per file and runs asynchronously. The MCP queues the job but doesn't wait for completion—you must poll the async result separately. Side-effect: adds media to your collection and consumes billing credits. Requires a valid HTTPS URL; some platforms block automated downloads.
@VideoDB ask my 'support-calls' collection: what are the top three customer complaints mentioned across all recordings this month?
Submitted grounded question to collection 'support-calls' using Search V2 indexes. Async operation ID: vdb_ask_5n8r4t1w This query will analyze indexed transcripts and return ranked themes. Typical response time: 30-90 seconds depending on collection size. The answer will cite specific video timestamps where complaints appear. This is an illustrative example; actual operation IDs and processing times vary by collection size and index configuration.
The 'Ask Collection' tool requires pre-built Search V2 indexes on your collection—ingestion alone isn't enough. This call is metered separately from ingestion. The MCP queues the question but returns immediately; you must poll for the grounded answer. Best for synthesis across many videos; single-video questions can use simpler search tools.
Use-case deep-dives
When VideoDB beats folder-and-transcript chaos for user research
A 5-person product team runs 8-12 customer interviews per sprint and needs to surface insights during planning without re-watching hours of footage. VideoDB wins here because the Ask Collection tool queries across indexed video transcripts in one call—no manual tagging, no Notion copy-paste graveyard. Ingest Media pulls recordings from Zoom or Loom URLs, and the async result pattern means the team can queue uploads in batch without blocking. The trade-off: if your interviews are under 10 minutes and you only review them once, plain transcripts in Notion are faster to set up. But once you hit 30+ videos or need to answer 'what did users say about onboarding friction' across six months of calls, the collection search justifies the API-key setup. Start with one sprint's interviews as a pilot collection.
How a 3-person support team indexes internal training footage
A startup support team has 40 internal training videos scattered across Google Drive and needs agents to find answers mid-ticket without watching full recordings. VideoDB's List Media and Get Media tools let the team catalog everything in one collection, then Ask Collection becomes the search bar agents actually use during live chats. The async ingestion model matters here—uploading 40 videos in one sitting won't time out the MCP session. The boundary: if your training content is already text (wiki pages, Notion docs), don't force it into video format just to use this MCP. But if the knowledge lives in screen recordings or demo walkthroughs and agents waste 15 minutes per ticket hunting clips, the indexed search pays back setup cost in week one. Create one collection, ingest the library, test search quality before rolling out.
When VideoDB makes sense for small marketing teams with video archives
A 4-person marketing team produces 6-10 video ads per quarter and needs to repurpose b-roll or find past campaign hooks without scrolling Dropbox folders. VideoDB's collection model groups assets by campaign or theme, and Ask Collection lets the team query 'show me all product demo footage from Q3' without manual tagging. The List Collections tool helps when you're managing multiple campaign buckets. The catch: if your team only produces 2-3 videos per year or never reuses footage, the API-key overhead and metered search costs outweigh the benefit—just use Dropbox search. But once you cross 20+ videos or spend more than an hour per month hunting clips, the indexed retrieval justifies the setup. Ingest your last two quarters as a test collection and measure search time saved.
Frequently asked
What does the VideoDB MCP let me do in Switchy?
It lets your team upload videos, audio, and images to VideoDB collections, then run semantic searches and ask questions across that media. You can create collections, ingest files from URLs, check async job status, and retrieve metadata — all without leaving your Switchy workspace. The MCP wraps VideoDB's API so you don't write code.
Do I need a VideoDB account to use this MCP?
Yes. You need a VideoDB API key, which you paste into Switchy's connection flow. The key determines which collections you can access and how much credit you have for metered operations like ingestion and search. If your team shares one VideoDB account, share the key; if you want separate billing, each person connects their own.
Can the MCP transcribe or edit video files?
No. It ingests media and runs semantic search or question-answering over collections, but it doesn't transcode, trim, or generate transcripts directly. If you need a transcript, you'd ingest the file, then use VideoDB's search features to extract spoken content. For editing, use VideoDB's web UI or a dedicated video tool.
How is this different from uploading files to VideoDB's dashboard?
The MCP lets you ingest and search from inside Switchy, so your team can chain VideoDB queries with other tools — pull a Notion doc, upload a video, ask a question, post the answer to Slack — in one workflow. The dashboard is faster for one-off uploads, but the MCP wins when you're automating or combining steps.
Who on my team should connect the VideoDB MCP?
Whoever owns your VideoDB account and has the API key. Ingestion and search are metered, so you probably want one shared connection to avoid duplicate billing. If your team runs separate projects with different collections, each project lead can connect their own key and stay within their credit budget.