The important moment isn’t when the analysis request is accepted. It’s when we can read the finished evidence and decide what it supports. A cooking instructor’s public knife-skills lesson is our example. This workflow explains the saved-analysis path, which is different from a transcript data job.
Kabo, an AI agent for creator research, offers a saved-analysis tool path. The client must keep its returned ID and read the final result before producing an editing memo.
Choose the client and confirm the connection
Connect https://kabo.sh/mcp-pub in a compatible client and complete Kabo sign-in. The MCP setup guide covers the connection; installation options are available for Codex and Claude Code.
The worked client is Codex. In the applicable config.toml, a direct connection uses [mcp_servers.kabo] with url = "https://kabo.sh/mcp-pub". Complete sign-in with the client’s MCP flow (codex mcp login kabo in the CLI), then inspect the available analysis tools.
[mcp_servers.kabo] url = "https://kabo.sh/mcp-pub"
Prepare the materials for this MCP task
[ ] Public source URL — The exact supported video we want to analyze. [ ] Analysis question — One issue such as explanation order or missing proof. [ ] Client connection — In this example, Codex with Kabo’s remote MCP server. [ ] Run record — The real returned analysis ID and current status. [ ] Owned footage — Clips we can use if the memo proposes a change.
Connect and propose the specific analysis
A cooking instructor uses a file-enabled Codex task to inspect a public knife-skills lesson they own. They want an editing memo about the explanation sequence, with a reliable record of whether the video analysis actually completed.
Configure the direct remote server in the applicable Codex configuration, complete Kabo authorization and inspect current tools. Ask the assistant to explain what it will analyze and why. New saved Kabo analyses require Pro and user confirmation; the connection alone is not the approval.
Inspect the Kabo analysis tools. I want to analyze [public URL] to answer [question]. Explain the proposed scope and that a new saved analysis requires Kabo Pro. Wait for my confirmation before submitting. If I supplied an existing analysis ID, inspect that result first.
Start once and retain the returned identifier
Call analyze_content_video with the supported public URL and a valid idempotency key. Use a new key for a new analysis; reuse it only to retry the same request. Preserve the returned id. The submission record is part of the workflow state, not yet an editing report.
The record the client should maintain
This is the sequence to keep in the task record. The placeholders are instructions, not a fabricated analysis result:
- New analysis: source_url + a new idempotency_key
- Returned id: Preserve the real id from the tool.
- Read result: get_analysis_run with analysis_id
- Pending: Continue status checks as appropriate.
- Failed / cancelled: Report the terminal outcome.
- Completed: Write only observations supported by the returned evidence.
Read the actual final state
Use get_analysis_run with analysis_id. Pending work remains pending. Failed and cancelled are final outcomes even when the tool response is not itself an error. Do not submit another analysis simply because the first is taking time. Inspect its state before deciding what to do next.
If a run is still pending, keep its ID. Starting another one doesn’t make the first result arrive sooner, and leaves us with two states to reconcile.
A transcript data job follows a different reader sequence. Keep its job_id separate from a saved analysis_id.
After confirmation, call analyze_content_video using the current schema, the public source_url and a valid new idempotency_key. Retain the returned ID; reuse the key only for a retry of this same request. Read get_analysis_run with analysis_id until final. Treat failed or cancelled as final outcomes. Only after completion, write an editing memo with supported observations, timestamps where available, proposed cuts and footage gaps.
Translate completed evidence into an editing memo
Extract supported observations about promise, explanation order and proof. Keep spoken and visual evidence distinct and use timestamps only when available. In the cooking example, a useful proposal could be to show an owned demonstration earlier, but only if the completed evidence supports the diagnosis and the footage exists.
Review the memo against the original footage
Save analysis-notes.md only if the client has authorized file access; otherwise return the text. Check every proposed cut in context. This tool path does not edit your timeline or establish private retention effects, and the analysis allowance rules differ from ordinary data retrieval.
The run is marked failed. Report that outcome, keep its identifier and separate any existing transcript observations from the unavailable visual analysis.
Approve the memo only after the run is complete
Use the approved memo in DaVinci Resolve with your original footage. Verify the sequence around every proposed cut and preserve the complete demonstration before considering titles or visual polish.
If the completed analysis changes the concept rather than a cut, take that finding into the idea brief. Start a different reference search only when the existing source cannot answer the question.
Worked examples are illustrative. Product screenshots show existing Kabo web results, not a run of the example in this guide. Use your own materials and the evidence returned in your task.