Clipto's $250M: a local index your agent can query

Clipto raised $15 million at a $250 million post-money valuation on August 31 for a product that searches your video, audio, images and documents in plain language. The funding is not the interesting part. The architecture is: the index runs on your machine, and an MCP server exposes it to your agent — so the model gets an answer without your archive getting a copy.

What actually happened

TechCrunch reported the all-equity round, backed by HSG, GL Ventures, EnvisionX Capital, Palm Drive Capital, 522 Ventures and angels including Hans Tung and Lu Zhang. Founded in 2023 by Henry Kang, the company indexes local media and lets you find things by description rather than filename. It recently added Model Context Protocol support so tools like Claude and ChatGPT can query the index, with processing running on the user's device.

The business numbers are the part worth writing down. Over 30 million people have used the product. Hundreds of thousands pay. Annual recurring revenue was around $15 million as of early 2026, the company is profitable on a net-income basis, and it does all of that with roughly 20 employees across the Bay Area, Hong Kong and Singapore. Its stated customer mix runs well past video creators into lawyers, doctors, researchers and professors.

Why local-first plus MCP matters for your business

Every AI search vendor pitching you right now defaults to the same shape: upload everything, we'll index it, query from our cloud. That is a fine deal until you look at what you just handed over — client files, case notes, patient recordings, unreleased product footage — and at your own contracts, which may not permit it.

Clipto's design inverts the default. The heavy work happens locally. MCP is the seam: a documented protocol that lets any compliant agent call your index as a tool instead of ingesting your files. That distinction is worth internalizing whether or not you ever buy this product. Your data does not have to move for a model to be useful on it. An MCP server in front of a local index, your own database, or your ERP gives an agent a narrow, logged, revocable door — and leaves the source of truth where it already lives, under your control.

The second lesson is commercial. Twenty people, $15 million ARR, profitable, no reliance on a hyperscaler's inference bill for the core loop. That is what a durable software business looks like when the expensive part runs on the customer's hardware instead of yours. If you are scoping an internal tool, price both shapes before you default to the cloud one.

Sources: TechCrunch — Clipto uses AI to search terabytes of video and is now valued at $250M.