Public API and MCP connector: use echoflow from Claude, Codex or your own systems

Your system uploads overnight, the content is there in the morning. Or ask your AI assistant to search years of video — without opening echoflow.

Try it free, no card

14 days · 60 minutes · 10 videos · no card

echoflow can be used without opening it: an MCP connector, included in every plan, turns it into a tool Claude, Claude Code, ChatGPT, Codex and other compatible assistants can call; with the Pro plan there's also a REST API to integrate it into your systems. Two routes to the same thing, meant for two different people on the same team.

The API, for developers

An agency with forty videos a week doesn't upload them by hand forty times. With the API your system:

pieces go one at a time and on interruption you resume from the last confirmed offset, exactly as the web app does;

Every key has its own rate limit and its own scopes: read for reading content, write for creating jobs, library for querying the archive. A key that only downloads shouldn't be able to spend your minutes.

The MCP connector, for everyone else

Connected once from the assistant, echoflow becomes a tool Claude, Claude Code, ChatGPT and Codex can use. From there you work in plain language: "take the latest episode, pull out the LinkedIn posts and tell me when I talked about pricing".

The assistant can process a video, read the content produced, search the archive and ask questions about your videos. That's the most interesting part: semantic search across years of content becomes available inside the tool you already work in: that, and the API, is where passage search lives.

You connect it with your own account, no keys: when connecting you pick the workspace and what to grant — read, create and process, query the archive — and you can revoke it whenever you like. Every request follows your plan as in the web app. The step-by-step guide is on the Connectors page. For an assistant without OAuth sign-in you use an API key instead, with the same scopes: that's on the Pro plan.

What we don't do

because we don't store social tokens. The API returns content, not publications.

other.

programmatic channel. Past the ceiling you get a clear error with a code, not a 500.

Which plan

The MCP connector with sign-in from your own account is included in every plan, trial included, at no extra cost. Public API, webhooks, chunked upload via API and API keys (also for connecting an assistant to the MCP server) are included in the Pro plan, in both sizes (€19/month for 200 minutes, €49 for 1,000 and up to 10 people). For volumes or terms beyond that there's Enterprise, on request.

Frequently asked questions

What is an MCP server, in practice?

It's how an AI assistant can use an external tool. With echoflow connected, you can ask Claude or Codex «take the latest episode, pull out the LinkedIn posts and tell me when I talked about pricing», and the assistant does it: processes the video, reads the outputs, searches the archive. Without opening our application.

Do I need to know how to code?

For the API yes, it's a developer integration. For the MCP connector no: you connect it once from the assistant with your own account (the «Connect» button, no keys) and work in plain language from there. They're two routes to the same thing, meant for two different people on the same team.

How do key scopes work?

Every API key carries its own scopes: «read» for reading produced content, «write» for creating jobs, «library» for querying the archive. They exist so you don't give an automation more power than it needs — a key that only downloads outputs shouldn't be able to spend your minutes.

Can I upload very large files via the API?

Yes, with chunked upload: you open an upload, send the pieces one at a time, and on interruption resume from the last confirmed offset. It's the same mechanism as the web app. The same plan limits apply as on the normal channel: past the ceiling you get a clear error, not a technical server failure.

Does the API consume minutes?

Yes, exactly like a manual upload: minutes are consumed by processing videos, whatever the channel. Querying the archive and reading outputs doesn't consume minutes.