Automatic Video Face Detection & Blur MCP, Restore Original — NoonAI DIS

Most tools target distinct resource+action pairs, and the descriptions explicitly draw the line between the one-shot convenience tools (encrypt_image/encrypt_video) and the split flow (upload_file + submit_encrypt_job + get_job_status). The closest overlaps are get_job_result vs download_job_result (metadata vs file bytes) and the four read-only billing/policy tools, which are separable but require reading descriptions carefully. purchase_credits vs purchase_credits_with_card and upload_file vs upload_file_content are clearly distinguished by storage mode.

All names are snake_case with a leading verb (get_, upload_, submit_, download_, purchase_, register_, encrypt_), which is predictable and readable. The minor deviation is encrypt_image/encrypt_video, which use 'encrypt' as the verb even though they actually perform upload+submit, making them look like a different family than submit_encrypt_job. Otherwise the pattern holds throughout.

At 15 tools the surface sits at the top of the comfortable range and is justified by two real subsystems: media anonymization and billing/payments. The billing side is slightly heavy with four read-only informational tools (billing status, pricing policy, usage summary, encrypt constraints) plus three mutation tools, but none is obviously redundant. Nothing is missing that would force an agent to fall back on out-of-band calls.

The core lifecycle is covered end-to-end: upload (path or base64), submit or one-shot encrypt for image and video, poll status, fetch result metadata, and download the output, plus constraints and full billing/payment management. Gaps are modest but real: there is no cancel/delete job, no filtered job listing beyond raw ULIDs from get_usage_summary, and no asset cleanup, so agents cannot abort or tidy failed runs.