French-American AI startup Reduck has raised $3 million in a pre-seed round, Tech.eu reports. Daphni led the round, with Drysdale Ventures and Galion.exe participating. The founders are Daniel Huynh (CEO), Louis Abraham (CTO) and Raphaël Millet (COO).
The problem it targets
Many websites and internal business tools offer neither an API nor a Model Context Protocol (MCP) interface. An AI agent that needs to use them has to work through the browser, clicking and typing its way through each step. According to the report, that approach can be slow, costly and unreliable on complex tasks.
What Reduck built
Reduck MCP is a platform where agents can discover, run and create browser automation scripts. An agent uses a predefined script instead of navigating a site step by step, which should cut the time and computing resources needed.
The company says this makes agents three times faster and cheaper on tasks such as downloading invoices from multiple vendors and doing market research on LinkedIn. That figure comes from Reduck; the article does not describe independent testing.
Scale and next steps
- The library holds more than 1,200 ready-to-run scripts covering over 250 websites.
- Reduck is developing a system of AI agents that create and maintain the scripts on their own, aiming for reliability and scalability.
- A developer offering is being prepared for teams that want custom agent systems connected to applications without APIs. The examples given are specialised insurance and healthcare portals and internal CRM systems. No release date or pricing is given.
- The company is also exploring automation of knowledge work, starting with generative engine optimisation (GEO).
The funding has supported hiring across AI engineering, full-stack development and infrastructure. Reduck plans to improve the technology, grow the script library and launch the developer product.
What is not yet known
The source gives no pricing, no customer names and no timing for the developer offering. How well the scripts hold up when a website changes is not addressed beyond the plan for agents to maintain them.