Automatic Video People De-identification MCP, Reversible Restore — NoonAI DIS

The image/video tools (encrypt_image, encrypt_video) substantially overlap with the split flow (upload_file/upload_file_content + submit_encrypt_job); the descriptions argue for the one-shot variants but an agent can easily pick either path. The three job tools (get_job_status, get_job_result, download_job_result) are separable but adjacent, and purchase_credits vs purchase_credits_with_card differ only in payment source. upload_file vs upload_file_content is the one overlap that is clearly resolved by the path-vs-base64 distinction.

All 15 tools use consistent snake_case with a leading verb (get_, upload_, submit_, encrypt_, purchase_, register_, download_). The pattern is predictable and the resource/action is readable from the name in every case.

Fifteen tools is reasonable for a pipeline that spans upload, submission, polling, retrieval, policy, and billing. It sits at the high end of well-scoped, with five billing/pricing tools being the heaviest cluster relative to the core de-identification work.

The full anonymization lifecycle is covered end to end: two upload paths, one-shot and split submission, status polling, result metadata, and file download, plus billing and constraint discovery. Gaps are minor: no job cancellation, no asset deletion or listing of assets, and job enumeration is only indirectly available through get_usage_summary.